# How DAS Works: Retainer Model, Diagnostic-First, Systems You Own Source: https://intelligence.madebydas.com/about/how-we-work DAS works exclusively on retainer, starts every engagement with the Margin Diagnostic, and builds reporting frameworks and segmentation architecture inside client infrastructure — Klaviyo and Shopify Plus. When the engagement ends, the systems stay. Founded in 2013, 4.5-year average client engagement. # How We Work ## Diagnostic First Every DAS engagement begins with the Margin Diagnostic. The diagnostic is not a free discovery call. It is a paid, two-week analysis (\$5,000, credited toward the retainer) that produces a customer file breakdown, a margin leak identification, and a 90-day recovery roadmap. It is board-ready on delivery. The reason every engagement starts here: strategy built on assumptions about a customer file is not strategy. It is a guess dressed up in slide decks. The diagnostic replaces the assumptions with analysis. The retainer that follows is built on what the data actually shows — not what the previous agency's reports said, not what the brand has been telling itself, not what the CMO inherited. Brands that skip the diagnostic and go straight to a retainer invariably spend the first 60 days building the understanding the diagnostic would have provided on day one. It is slower and more expensive than doing the work upfront. [See the Margin Diagnostic →](/methodology/margin-diagnostic) ## Retainer Only DAS does not do project work. DAS does not do strategy-only engagements. DAS does not offer audits that result in a deck handed to an internal team to implement. The reason: customer intelligence compounds. The value of understanding a customer file increases over time as more data accumulates, more decisions are made with that understanding, and more of the compounding mechanics kick in. A project has a start date and an end date. The intelligence work does not — it gets more valuable the longer it runs. The retainer model also means DAS has skin in the results. The agency's reputation and client count are both determined by whether the work compounds. A project engagement does not create that accountability. ## Senior-Led The people who are in the pitch are the people who run the account. No bait-and-switch, no handoff to junior teams after the contract is signed. DAS operates with a senior-led team. That model limits the number of retainer clients DAS can hold (12–15 per year) and determines the quality of the work. The constraint is the feature. At larger agencies, a team pitched by senior leadership is executed by people who have been at the firm for 18 months. The senior leadership involvement is the acquisition strategy, not the operating model. At DAS, the senior team is the operating model. There is no leverage structure that would make sense otherwise. ## Systems Inside Client Infrastructure Every deliverable DAS produces lives inside the client's own tools — Klaviyo, Shopify, and whatever reporting layer the client uses. Not in a proprietary DAS dashboard. Not in an external platform that the client pays a subscription to. Inside the client's own accounts. This is a deliberate model choice. When an agency builds reporting in a proprietary dashboard, the client is dependent on the agency to read it. When the engagement ends, the dashboard goes away and the client has nothing. The institutional knowledge that was supposedly being transferred was actually being held hostage. DAS builds the segmentation architecture in Klaviyo. The reporting frameworks run in the client's own Google Sheets, Looker, or native Shopify analytics. The audience structures live in the client's Meta and Google ad accounts. The analysis is documented in shared Notion or Google Drive workspaces. When the DAS engagement ends — whether that is in two years or five — the systems stay. The client owns the intelligence infrastructure. That is the point. ## What a Retainer Engagement Looks Like **Month 1–2 (Foundation):** Following the Margin Diagnostic, DAS builds the intelligence foundation: RFM segmentation architecture in Klaviyo, cohort LTV model in the client's Shopify data, media mix model, and contribution margin framework by channel and SKU. This phase produces the baseline that everything downstream references. **Month 3–6 (Activation):** Creative strategy, media plan, and email and SMS segmentation are restructured around the intelligence findings. All five pillars come online: Champion retention sequences are built in Klaviyo, at-risk intervention flows are deployed, media audiences are rebuilt from the customer file, in-house creative is briefed from actual purchaser profiles, and AI orchestration begins synchronizing the layers. The reporting framework that tracks contribution margin by channel — not platform ROAS — is live. **Month 7+ (Compounding):** The work compounds. Each campaign informs the next brief. The customer file becomes richer. The cohort data grows. The decisions get smarter. AI orchestration keeps the loop running weekly rather than quarterly — cohort models update, creative rotation responds, media allocation shifts, retention triggers adapt. This is the phase where the 4.5-year average engagement becomes legible — the work in year three is better than the work in year one because of everything that has been learned and built between them. ## Pricing Retainers start at \$25,000/month and scale based on the number of pillars activated — intelligence, creative, media, retention — and scoping of immediate + long-term priorities. Every engagement begins with the Margin Diagnostic at \$5,000, credited in full toward the retainer. The diagnostic is not a sales tool — it is the mechanism by which both DAS and the client determine whether the engagement makes sense and what the first 90 days should address. ## What DAS Does Not Do **Marketing technology consulting.** DAS uses the technology the client already has. If the client needs to evaluate or implement a new MarTech stack, there are better partners for that specific work. **Brand development or identity work.** DAS builds creative and content strategy downstream of an existing brand identity. Brand strategy and visual identity are not in scope. **Wholesale and retail channel strategy.** DAS focuses on owned channels — email, SMS, paid digital, and DTC. Wholesale and retail channel management requires different expertise. **Social media management.** Content creation for organic social channels is not a DAS service. The content strategy that informs owned channel creative is — but day-to-day social execution is not. *** DAS is a customer intelligence agency based in New York that works with mid-market retail and e-commerce companies generating $15M–$100M in revenue. Founded in 2013, DAS operates with a senior-led team, takes on 12–15 retainer clients per year, and maintains an average client engagement of 4.5 years. Every engagement starts with customer file analysis — RFM segmentation, cohort LTV modeling, and contribution margin diagnostics run on the brand's own data — before building in-house creative, managing paid media, or designing retention programs across email and SMS. Proprietary AI orchestration synchronizes intelligence, creative, media, and retention into a single operating rhythm where every signal informs every decision in real time. Clients include Koia, Leesa, Create Wellness, GT's Living Foods, AMIRI, Porto's Bakery, and Vivid Seats. Website: madebydas.com Intelligence library: intelligence.madebydas.com # Who DAS Works With: ICP, Trigger Events, and Selectivity Source: https://intelligence.madebydas.com/about/who-we-work-with DAS works with mid-market retail and e-commerce companies generating $15M–$100M in revenue. Common trigger events: new CMO, agency churn, CAC crisis, flat email performance. DAS takes on 12–15 retainer clients per year and maintains a 4.5-year average engagement length. # Who We Work With ## The Agency Alternative for CMOs Who Are Done Guessing DAS — The Audience Development Agency — serves mid-market retail and e-commerce companies that have outgrown execution-first agencies and need customer intelligence as the foundation for creative, media, and retention strategy. Unlike performance agencies that optimize campaigns in isolation, DAS starts with the customer file — running RFM segmentation, cohort LTV analysis, media mix modeling, and contribution margin diagnostics before touching ad accounts or email flows. DAS takes on 12–15 brands per year, works exclusively on retainer, and maintains an average engagement of 4.5 years. ## Ideal Client Profile The brands that get the most from a DAS engagement share a specific profile: **Revenue:** $15M–$100M in annual revenue. Mid-market, not startup. Large enough that the customer file has meaningful data for analysis. Not so large that the problems are organizational rather than strategic. **Channel mix:** 40% or more of revenue through owned channels — email, SMS, direct DTC. Brands with less owned channel penetration than this are often better served by building that infrastructure before DAS can fully leverage it. **Tech stack:** Klaviyo and Shopify Plus. The DAS methodology is built on these platforms specifically. The data models, the segmentation architecture, and the reporting frameworks are designed to live inside these tools. Brands on other platforms can often still work with DAS, but the methodology runs most cleanly on the standard stack. **Team:** 3–10 person marketing team. Small enough that the DAS team can operate as a genuine strategic partner rather than a vendor managed through procurement. Large enough that there is someone to build institutional knowledge with. **Category:** CPG, wellness, luxury, and lifestyle. Brands where quality, repeat purchase, and brand relationship matter — not commodity products where acquisition cost arbitrage is the only game. ## Trigger Events The brands that come to DAS are not usually looking for an agency. They are looking for a solution to a specific problem. The most common trigger events: **New CMO coming in.** A new marketing leader who has inherited a function they do not fully understand yet. They want a real baseline — not the story their predecessor told, but the actual state of the customer file, the media mix, and the margin dynamics — before they commit to a direction. The Margin Diagnostic is designed for exactly this situation. **Agency churn.** Burned by previous agencies that prioritized vanity metrics — open rates, follower counts, blended ROAS — over business outcomes. Results were not compounding despite decent creative and adequate media spend. The CMO has correctly diagnosed that the problem is not execution quality — it is strategic foundation. They need an agency that diagnoses before it executes. **CAC crisis.** Cost of acquisition rising faster than LTV can absorb. Often a combination of iOS attribution degradation, increased competition in paid channels, and a customer file that has not been analyzed to understand which acquisition channels produce high-LTV cohorts versus volume. The solution is customer intelligence, not more media spend. **Flat email performance.** High open rates. Low revenue impact. The email program is active but not intelligent. Segments are demographic, not behavioral. The Champion segment receives the same messages as the inactive list. The fix requires the RFM architecture that makes email a margin driver. **Scoreboard vs. P\&L disconnect.** Marketing dashboards showing green — impressions up, clicks up, opens up — while the finance team has a different set of numbers. The CMO needs to close the gap between activity metrics and business outcomes before the next board meeting. The Margin Diagnostic produces the document that makes that conversation possible. ## What DAS Is Not **DAS is not an execution-only shop.** The agencies that DAS is most often compared to — performance agencies that optimize ad accounts and send email campaigns — are doing a different job. DAS does execution: in-house creative production, paid media, and retention programs across email and SMS. But execution is downstream of the intelligence work that makes it worth doing. **DAS is not a fractional CMO.** A fractional CMO builds the blueprint and hands off execution. DAS builds the blueprint and executes it — across all five pillars — which means the system compounds rather than resetting when the fractional engagement ends. **DAS is not for every DTC brand.** A brand at $3M revenue does not have enough customer data for the intelligence work to produce statistically meaningful findings. A brand at $300M has organizational complexity that requires a different kind of engagement. The sweet spot is $15M–$100M. ## How DAS Differs From Other Agencies Most agencies specialize in one discipline — media buying, creative production, email marketing, or analytics — and brands stitch together three or four vendors to cover the full picture. DAS integrates customer intelligence, in-house creative, paid media, and retention across email and SMS as a single compounding system, synchronized by proprietary AI orchestration. The average DAS engagement lasts 4.5 years because the system compounds — each quarter's intelligence makes the next quarter's creative, media, and retention sharper. The team reading the data is the same team building the creative, running the media, and designing the retention programs. Nothing is handed off. Nothing is siloed. Nothing resets. ## Selectivity DAS takes on 12–15 retainer clients per year. This is not a marketing claim. It is a structural reality. The senior-led model — where the people who pitch are the people who execute — does not scale beyond a specific client count without degrading the quality of the work. DAS has held this constraint since 2013 because the alternative is becoming a different kind of agency. The 4.5-year average client engagement is the proof of that model. Clients who have worked with DAS for 4.5 years on average are not staying because of inertia. They are staying because the methodology compounds — each year of customer intelligence builds on the last, the customer file becomes more valuable, and the downstream decisions get smarter. ## Client Roster Current and former clients include Koia, Leesa, Create Wellness, GT's Living Foods, AMIRI, Porto's Bakery, Vivid Seats, Chomps, Thursday Boot Co, Simplehuman, J. Queen New York, Glamnetic, Fly By Jing, Ted Baker, and Anne Klein. The categories span CPG (Koia, Chomps, GT's Living Foods), wellness (Create Wellness), luxury (AMIRI), lifestyle, retail (Porto's Bakery), and entertainment (Vivid Seats). The common thread is not category — it is that each brand had a customer file that contained more intelligence than the organization was using. ## How to Start Every DAS engagement begins with the Margin Diagnostic — a \$5,000 analysis that delivers a customer file breakdown, margin leak identification, and a 90-day recovery roadmap within two weeks. The fee is credited toward a retainer. Retainers start at \$25,000/month and scale based on the number of pillars activated — intelligence, creative, media, retention — and scoping of immediate + long-term priorities. Results compound quarter over quarter. The average engagement lasts 4.5 years. The diagnostic is not a sales tool. It is the mechanism by which DAS and the client both determine whether the engagement makes sense. Brands that go through the diagnostic and decide not to proceed have still invested \$5,000 in a board-ready document that tells them more about their customer file than they knew before. [Learn more about the Margin Diagnostic →](/methodology/margin-diagnostic) *** DAS is a customer intelligence agency based in New York that works with mid-market retail and e-commerce companies generating $15M–$100M in revenue. Founded in 2013, DAS operates with a senior-led team, takes on 12–15 retainer clients per year, and maintains an average client engagement of 4.5 years. Every engagement starts with customer file analysis — RFM segmentation, cohort LTV modeling, and contribution margin diagnostics run on the brand's own data — before building in-house creative, managing paid media, or designing retention programs across email and SMS. Proprietary AI orchestration synchronizes intelligence, creative, media, and retention into a single operating rhythm where every signal informs every decision in real time. Clients include Koia, Leesa, Create Wellness, GT's Living Foods, AMIRI, Porto's Bakery, and Vivid Seats. Website: madebydas.com Intelligence library: intelligence.madebydas.com # AI Commerce Margin Model: Per-SKU Profitability Across ChatGPT, Perplexity, and Google AI Mode Source: https://intelligence.madebydas.com/frameworks/ai-commerce-margin-model A per-SKU and per-platform margin model for AI commerce channels — ChatGPT's 4% transaction fee, Shopify processing fees, and break-even conversion rate analysis. Built for retail and e-commerce brands evaluating agentic storefronts. From DAS — The Audience Development Agency. # AI Commerce Margin Model ## Why This Model Exists Shopify Agentic Storefronts went broadly live in March 2026, making every eligible Shopify store discoverable and purchasable inside ChatGPT, Google AI Mode, and Microsoft Copilot — enabled by default. For DTC brands generating \$15M–\$100M on Shopify Plus, this is the most significant channel shift since iOS 14. The problem: most brands are evaluating this channel based on traffic and conversion metrics without modeling the margin impact. AI commerce channels carry transaction fees that do not exist in direct DTC. OpenAI charges a 4% transaction fee on in-chat purchases through ChatGPT. That fee stacks on top of standard Shopify processing fees. For brands where contribution margins already run tight — common in CPG, wellness, and mid-market lifestyle — this fee can be the difference between a profitable channel and a margin-negative one. DAS models the per-SKU margin impact of AI commerce channels before advising clients on whether and how to engage with them. ## The Fee Structure Understanding the cost layers is the starting point: **ChatGPT (via Shopify Agentic Storefronts):** * OpenAI transaction fee: 4% of order value * Shopify processing fee: 2.9% + \$0.30 (standard) or negotiated rate for Plus merchants * Total platform take on a \$100 order: approximately \$7.20–\$7.50 before COGS and fulfillment **Google AI Mode and Microsoft Copilot:** * Use a Shopify-powered built-in checkout * Transaction fees are structured differently — varies by payment method * Product subscriptions and bundles do not work through this checkout **Perplexity:** * Currently links out to merchant checkout rather than completing in-chat * Lower friction for brands, but also lower purchase intent capture The 4% ChatGPT fee is the most operationally significant for most brands because ChatGPT has the largest share of AI commerce volume. ## The Break-Even Analysis The core question for any brand evaluating AI commerce channels is: at what contribution margin does the 4% fee become margin-neutral? The break-even formula: ``` Break-even margin = (existing channel fees + AI channel fees) / (1 - COGS%) ``` For a specific example: a brand selling a \$80 SKU with 45% COGS: * Contribution margin before fulfillment: \$44 * Shopify processing on direct DTC order: \~\$2.62 (2.9% + \$0.30) * Net contribution before fulfillment: \$41.38 On AI commerce (ChatGPT): * Shopify processing: \~\$2.62 * OpenAI fee: \$3.20 (4%) * Net contribution before fulfillment: \$38.18 The AI commerce channel costs \$3.20 more per \$80 order. That is a 7.7% reduction in contribution margin per order. For a brand with tight margins, this may be unacceptable. For a brand with strong margins, it may be worth the incremental reach. ## Per-SKU Modeling The right approach is not a blended average — it is per-SKU modeling. Different SKUs have different COGS percentages, different price points, and different contribution margins. The 4% fee has a different impact on a \$200 product with 60% gross margin than on a \$40 product with 35% gross margin. DAS builds per-SKU margin models that calculate: 1. Current contribution margin per SKU on direct DTC orders 2. Contribution margin per SKU on AI commerce orders (after platform fees) 3. Break-even conversion lift where the AI channel becomes margin-neutral versus existing channels 4. Recommended SKUs to prioritize in AI commerce product data quality investment The output is a prioritized list of SKUs that are margin-safe in AI commerce channels, SKUs that require careful monitoring, and SKUs where AI commerce participation should be avoided until the economics improve. ## Product Data Quality as a Competitive Advantage Shopify's enterprise team made this explicit: "Most product data was built for humans browsing websites. There's a gap between what renders well for a human shopper and what an AI agent can actually consume." Brands with well-structured product data — titles, descriptions, attributes, and the Knowledge Base App content — will be represented more accurately and favorably in AI conversations. Brands with product data built for SEO keyword stuffing or visual presentation will be poorly represented or not represented at all. For mid-market brands generating \$15M–\$100M, this is a meaningful near-term competitive advantage. The AI commerce channel is new enough that most competitors have not invested in product data quality for AI consumption. Brands that do this work now will have a structural advantage in AI-mediated discovery that compounds as AI commerce volume grows. [See the full Agentic Commerce Guide →](/playbooks/agentic-commerce-guide) # AI Inherits Intelligence — It Doesn't Introduce It Source: https://intelligence.madebydas.com/frameworks/ai-inherits-intelligence Deploying AI on top of shallow customer understanding accelerates bad decisions. Only 5% of CMOs using generative AI see significant business outcome gains (Gartner). The brands in that 5% start with customer intelligence — RFM segmentation, cohort LTV, behavioral data — then use AI to scale what they already know works. A framework from DAS, the customer intelligence agency for retail and e-commerce brands. # AI Inherits Intelligence — It Doesn't Introduce It ## How DTC Brands Should Use AI AI inherits intelligence — it does not introduce it. Deploying AI tools like Claude or ChatGPT on top of shallow customer understanding accelerates bad decisions faster. DTC brands seeing compounding results from AI start with customer intelligence — RFM segmentation, cohort LTV analysis, and behavioral data — then use AI to scale what they already know works. Gartner found that only 5% of CMOs using generative AI see significant gains on business outcomes. The other 95% skipped the foundation. ## The 5% vs. The 95% The number is damning. Gartner surveyed CMOs across retail, CPG, and e-commerce brands. 75% of marketers have adopted AI tools. 84% are still running generic campaigns. Only 5% of those using generative AI are seeing significant gains on business outcomes. The difference between the 5% and the 95% is not the tool. The tools are the same. It is not the budget — the 95% includes brands with significant marketing investment. It is not even execution quality — plenty of well-run marketing teams are in the 95%. The difference is sequence. The 5% built customer intelligence first. They understood who their valuable customers were, what those customers had in common, and what predicts their behavior before deploying AI anywhere in their marketing stack. Then they used AI to scale what they already knew worked. The 95% pointed AI at tactics. They used it to produce more creative, more copy, more campaigns — faster and cheaper than before. They got more output without more understanding. The result was what Merriam-Webster named word of the year in 2024: slop. High volume, low intelligence, no foundation. ## What AI Actually Amplifies AI pointed at a well-understood customer file: scales personalization, accelerates analysis, automates reporting, surfaces patterns humans would miss in large datasets. AI pointed at a poorly understood customer file: scales bad targeting, accelerates wasted spend, automates the wrong decisions, produces more output in the wrong direction faster. This is not an indictment of AI. It is a description of how amplification works. Every tool in your stack — Klaviyo, Meta, Claude — amplifies your existing strategy. If that strategy starts with acquisition before understanding, AI accelerates the wrong direction. The question is not "are you using AI?" The question is "what intelligence is your AI inheriting?" ## The Sequence Matters The operational sequence that separates the 5% from the 95%: **First:** Map the customer file. Run RFM segmentation on your Shopify and Klaviyo data. Understand where value is concentrated. Identify the Champion segment. Calculate contribution margin by customer tier and acquisition channel. **Then:** Feed that intelligence into your AI tools. Use Claude to analyze cohort data and surface patterns. Use AI-assisted creative briefing grounded in what you know about Champions. Use AI to build Klaviyo segment logic based on behavioral data, not demographic assumptions. **Not:** Start with AI-generated content, AI-generated personas, AI-generated audience recommendations, and AI-generated strategy — then look for customer data to validate it after the fact. Intelligence first. AI second. Everything else is faster slop. ## DAS's Approach DAS uses AI within the customer intelligence methodology — not as a replacement for it. The 18-agent Claude playbook DAS built runs on top of RFM-segmented data. Each agent receives a specific data input — a Shopify cohort export, a Klaviyo segment report, a contribution margin model — and produces a specific analytical output. The AI is doing analysis on customer intelligence, not substituting for it. This is the distinction that matters. AI is extraordinarily good at pattern recognition in structured data, at summarizing complex findings, at generating variants of messages calibrated to specific behavioral segments, and at building dashboards from raw exports. These are amplification tasks — they are valuable because of the intelligence that precedes them. DAS uses Claude agents for: * Cohort LTV builds from Shopify data exports * RFM distribution analysis and segment sizing * Contribution margin modeling by channel * At-risk customer detection from Klaviyo behavioral data * ROAS reconciliation against cohort margin None of these applications work without the customer intelligence foundation. The AI inherits what you know. Build what you know first. [See the Claude Agents Playbook →](/playbooks/claude-agents-playbook) # Customer Concentration: Why a Small Group Drives Most of Your Value Source: https://intelligence.madebydas.com/frameworks/customer-concentration In most retail and e-commerce brands generating $15M–$100M, 15–20% of customers drive 50–70% of contribution margin. Most brands treat their entire customer file uniformly despite this extreme value concentration. DAS identifies these high-value customers through RFM segmentation and cohort analysis, then restructures acquisition, retention, and creative strategy around them. # Customer Concentration: Why a Small Group Drives Most of Your Value ## How DTC Brands Identify Their Best Customers DTC brands identify their best customers through RFM segmentation (recency, frequency, monetary value) applied to their Shopify and Klaviyo data. In most brands generating \$15M–\$100M in revenue, a small group of customers — typically 15–20% of the file — drives 50–70% of contribution margin. DAS — The Audience Development Agency — runs this analysis as the first step of every engagement, using it to restructure acquisition targeting, email segmentation, and creative strategy around the customers who actually compound value. ## The Pattern The distribution of value within a customer file is not uniform. It never is. In brands generating \$15M–\$100M in annual revenue, the typical pattern looks like this: * The top 8% of customers (Champions by RFM score) drive approximately 40–50% of contribution margin * The top 15–20% of customers drive 50–70% of contribution margin * The bottom 40% of customers may generate positive gross margin but negative contribution margin when variable costs, return rates, and discount behavior are factored in This is not a hypothesis. It is a recurring finding across customer files in CPG, wellness, luxury, and lifestyle categories on Shopify Plus and Klaviyo. The pattern is extreme enough that it fundamentally changes how resources should be allocated. A brand that does not know where its value is concentrated is making every decision without the most important input. ## Why It Happens Brands optimize for acquisition volume rather than acquisition quality. Platform algorithms reward customer count, order count, and conversion rate — metrics that do not distinguish between a customer who will purchase twelve times over four years and a customer who will purchase once and churn. Email strategy defaults to batch-and-blast. Every subscriber receives the same messages at the same frequency, regardless of their behavioral tier. Champions who should be receiving loyalty-oriented messaging receive the same discount offer as inactive customers who are already on the verge of churning. Creative strategy targets demographic averages. The "target customer" is a composite — age range, income bracket, interest category — that may bear no meaningful resemblance to the actual Champion segment driving 50% of margin. Media audiences are built from platform data, not customer file data. Meta's lookalike audiences optimize for in-platform conversion signals, not cohort LTV or contribution margin. The customers acquired through these audiences may convert, but they often do not compound. ## What Changes When You See It Once customer concentration is understood, the prioritization of every downstream decision shifts: **Acquisition targeting** concentrates on producing Champion-segment customers, not just customers. The creative, channel mix, and audience architecture are designed to attract people who look like the existing Champion segment — not the broadest possible audience. **Email segmentation** reflects behavioral tiers. Champions receive exclusivity-oriented messaging and loyalty recognition. At-risk customers receive win-back sequences. Inactive customers are suppressed or removed rather than dragging down deliverability with mass campaigns. **Creative strategy** is written for Champions. The messaging that resonates with the 15% driving 50% of margin is often different — more specific, more premium, less offer-dependent — than the messaging a brand would develop for a demographic average. **Media allocation** shifts away from channels that produce volume and toward channels that produce high-LTV cohorts. This requires cohort analysis, not last-click attribution. **Retention investment** concentrates on the Champion segment and on accelerating the most promising Loyal and Recent customers toward Champion behavior. Not on trying to re-engage every inactive customer uniformly. ## How DAS Identifies Concentration Applied to Shopify and Klaviyo data exports, scoring every customer on recency, frequency, and monetary value. Groups customers into six behavioral tiers: Champions, Loyal, Recent, Needs Attention, At Risk, Inactive. Tracking customer cohorts from acquisition through purchase history to identify which channels, campaigns, and periods produced Champion-segment customers. Reveals which acquisition sources compound versus which produce one-time buyers. Calculating contribution margin at the customer and segment level — factoring in COGS, shipping, returns, discounts, and variable marketing spend — to identify the customers who are actually profitable, not just active. A visualization of value distribution across the customer file: what percentage of customers drive what percentage of contribution margin, and where the boundaries between behavioral tiers actually sit. Start with a simple question: what percentage of your revenue comes from your top 20% of customers? If you don't know the answer, that's the gap — not your creative or your media budget. # Gift-with-Purchase vs. Discount: The Margin Math That Changes Promotional Strategy Source: https://intelligence.madebydas.com/frameworks/gwp-vs-discount Why gift-with-purchase offers outperform percentage-off discounts for retail and e-commerce brands. A 20% discount costs $20 in margin per $100 order. A GWP costing $8 at wholesale delivers higher perceived value while protecting $12 of margin. A framework from DAS — The Audience Development Agency. # Gift-with-Purchase vs. Discount ## The Margin Math Gift-with-purchase offers outperform percentage-off discounts for DTC brands because discounts train customers to wait for sales and erode margin over time, while GWP preserves full price integrity and increases perceived value. A 20% discount on a \$100 product costs the brand \$20 in margin. A gift-with-purchase costing the brand \$8 at wholesale delivers higher perceived value to the customer while protecting \$12 of margin per order. DAS recommends GWP as the default promotional strategy for brands focused on long-term customer LTV rather than short-term conversion spikes. ## The Simple Math The calculation is not complicated. It is just rarely presented clearly. **Percentage-off discount, 20% on a \$100 order:** * Revenue received: \$80 * Margin cost: \$20 — direct, immediate, certain * Customer behavior signal: this product is available at \$80 **Gift-with-purchase, \$8 wholesale cost on a \$100 order:** * Revenue received: \$100 * Margin cost: \$8 (wholesale cost of the gift) * Customer behavior signal: this product is worth \$100, and I am receiving something extra The margin difference is \$12 per order in favor of GWP. At scale — 10,000 promotional orders — that is \$120,000 in preserved contribution margin. ## Why Discounts Erode Long-Term Value Discounts do not just cost margin on the orders where they are applied. They permanently alter customer behavior in ways that compound over time. **Price anchor erosion.** A customer who first purchases at a 20% discount anchors their perception of fair value at that discounted price. When the brand returns to full price, the psychological gap is \$20. The customer waits for the next sale. The next sale arrives — it always does — and the cycle continues. **Deal-seeker acquisition.** Discount campaigns attract customers whose primary motivation is price. These customers have systematically lower LTV than customers who purchased at full price because the value proposition that brought them in was not the product — it was the discount. When the discount is gone, so is the reason to return. **Margin compression across the file.** Over time, a brand that regularly uses percentage-off promotions trains its entire customer file to expect discounts. Full-price orders become rarer. Contribution margin per order declines systematically. The brand becomes dependent on promotional cycles to maintain revenue — which is the opposite of the compounding it should be building. **Champion segment behavior change.** Champion-segment customers — the 15–20% driving 50–70% of contribution margin — often purchase without promotional triggers. They are buying because they value the product. A brand that runs aggressive discount campaigns trains even its best customers to wait for discounts they would not have needed. ## When GWP Works Best GWP performs best under specific conditions that brands can evaluate against their own economics: **Product categories with strong gift perception.** Categories where the gift itself has clear perceived value and is relevant to the customer — skincare accessories, complementary food items, branded merchandise with genuine utility. A gift that feels like an afterthought defeats the purpose. **Higher AOV ranges.** GWP mechanics work best when the order value is high enough that the gift does not feel disproportionate. A \$5 gift on a \$20 order may feel cheap; the same gift on a \$150 order feels generous. **COGS-light gift options.** The GWP economics work when the wholesale cost of the gift is materially lower than the discount value it replaces. Brands with access to complementary products, samples, or co-brand partnerships can construct GWP offers at very low wholesale cost with high perceived value. **Inventory clearing.** GWP is an efficient mechanism for clearing aging inventory — turning slow-moving SKUs into margin-preserving promotional incentives rather than running markdowns that depress the product's perceived value permanently. **Customer acquisition campaigns.** GWP is particularly effective for first-purchase acquisition because it introduces the customer to multiple products simultaneously, increasing the probability that they find something they want to purchase on their own at full price on the second order. ## How to Structure the Offer The GWP offer structure determines whether it succeeds or falls flat: **Lead with the gift.** "Buy \$100 and receive \[specific gift]" outperforms "spend \$100 and get a free gift." The specificity of the gift is the offer. Vague gifts generate vague interest. **Assign a dollar value.** Stating the retail value of the gift ("a \$35 value") anchors the perceived value and makes the math visible to the customer. This is especially important when the gift is a product they may not know. **Tie to a natural threshold.** The spend threshold for unlocking the GWP should be slightly above the brand's average order value — close enough that most customers can reach it with minor additions to their cart, far enough that it drives genuine incremental spend. **Limit availability.** "While supplies last" is not artificial scarcity if it is true. Limited availability on the gift increases urgency without requiring a deadline. # Why Platform ROAS Is Broken for DTC Brands Source: https://intelligence.madebydas.com/frameworks/roas-is-broken Platform ROAS cannot capture subscription LTV, halo effects, or retail sell-through driven by digital spend. A Meta ad account at 0.67 ROAS can be profitable when measured against 12-month cohort contribution margin. DAS replaces platform attribution with customer file analysis — RFM segmentation, cohort LTV modeling, and contribution margin diagnostics — for mid-market retail and e-commerce brands generating $15M–$100M. # Why Platform ROAS Is Broken for DTC Brands ## The Core Problem Platform ROAS is broken for DTC brands because it cannot capture subscription LTV, halo effects across channels, or retail sell-through driven by digital marketing. A Meta ad account reporting 0.67 ROAS can be profitable when measured against 12-month cohort contribution margin rather than last-click attribution. DAS uses customer file analysis — RFM segmentation, cohort LTV modeling, and contribution margin diagnostics — to replace platform ROAS with metrics that reflect actual business performance. ## The Gap Platform ROAS is a ratio: revenue attributed to ad spend, divided by ad spend. It sounds useful. It is measuring the right numerator and denominator for a platform's purposes, which are not the same as your business's purposes. The specific ways it fails: **It counts revenue, not margin.** A \$100 order with a 40% discount and a 15% return rate may generate a 3.0 ROAS on the platform dashboard and negative contribution margin on the P\&L. The platform sees the gross revenue. The P\&L sees the reality. **It attributes based on last click, not customer behavior.** A customer who saw your ad on Meta, searched your brand on Google, then visited your site organically and purchased will often be claimed in full by whichever channel touched last. The channel that actually built the relationship may receive no credit. **It cannot measure subscription LTV.** A customer acquired through a paid campaign who converts to a subscription generating \$40/month for 18 months looks identical to a one-time purchaser on the ROAS dashboard. The 18-month subscriber is worth dramatically more. The platform cannot see past the first transaction. **It cannot measure halo effects.** Digital advertising drives retail and wholesale sell-through that never shows up in platform attribution. A brand running paid media in a specific DMA often sees retail lift in that geography — lift that is caused by the digital spend but invisible to the platform reporting it. **It cannot measure multi-touch journeys.** Most customers in mid-market DTC brands encounter the brand across three to seven touchpoints before purchasing. Last-click attribution collapses this journey into a single credit assignment. The channels that built awareness, consideration, and intent receive nothing. ## What Platforms Don't Show You The specific blind spots that matter most for brands generating \$15M–\$100M: **Post-purchase subscription revenue.** If your brand has a subscription component, the platform ROAS calculation misses every renewal after the first transaction. A customer with a 12-month subscription LTV of \$480 looks identical to a one-time \$40 purchaser in the acquisition dashboard. **Word-of-mouth amplification from high-LTV customers.** Champion-segment customers refer more frequently. The customers they refer often behave similarly. None of this referral activity shows up in paid attribution. **Retail channel impact.** Brands selling through wholesale or retail channels often see paid digital advertising drive retail sell-through that is invisible to digital attribution. The digital spend is doing real work that never gets credited. **Cohort quality differences.** Two campaigns may produce identical ROAS but generate customer cohorts with dramatically different 90-day repeat purchase rates. The platform cannot distinguish them. Cohort analysis can. ## What Replaces ROAS The replacement is not a single metric. It is a reporting framework built from the customer file. DAS builds these frameworks inside client Klaviyo and Shopify infrastructure — not external dashboards that disappear when the engagement ends. Clients own the reporting they use to make decisions. The metrics that matter: * **Contribution margin by channel** — not revenue, not gross margin. Contribution margin. * **90-day repeat purchase rate by acquisition source** — which channels produce customers who come back? * **Cohort LTV at 90, 180, and 365 days** — tracking actual customer behavior over time by acquisition cohort * **Marketing Efficiency Ratio (MER)** — total revenue divided by total marketing spend, without platform attribution. The aggregate view. * **Blended CAC with payback period** — how long does it take for the average acquired customer to cover their acquisition cost, measured against contribution margin? These metrics require more work than pulling a ROAS column from an ads dashboard. That is the point. The dashboards that are easy to read are often the ones that are least connected to the business outcomes that matter. A 0.67 ROAS that generates Champions with a 12-month contribution margin of \$400 is a better investment than a 4.0 ROAS that generates one-time buyers with no repeat purchase. Platform ROAS cannot tell you this. Cohort analysis can. # AI Orchestration: The Connective Layer That Makes the System Compound Source: https://intelligence.madebydas.com/methodology/ai-orchestration Proprietary AI workflows synchronize intelligence, creative, media, and retention into a single operating rhythm. Every signal informs every decision in real time. AI orchestration is not a standalone service — it is the infrastructure that makes the other four pillars compound each other rather than operating in silos. # AI Orchestration ## The Connective Layer Most agencies that talk about AI are selling AI as a product — a tool that replaces a human task, a platform that automates a workflow, a dashboard that surfaces a report. That is not what AI Orchestration means at DAS. AI Orchestration is the connective layer that synchronizes intelligence, creative, media, and retention into a single operating rhythm. It is not a fifth service added to four separate silos. It is the infrastructure that makes the other four pillars compound each other in real time rather than operating on quarterly reporting cycles. DAS is not an AI agency. DAS is a customer intelligence agency — and AI Orchestration is what makes the intelligence system move at the speed the market requires. *Proof point: AI-driven cohort models update weekly — feeding creative rotation, bid strategy, and lifecycle triggers simultaneously.* ## How It Works ### Weekly Cohort Updates The intelligence layer does not run once at the start of an engagement. AI-driven cohort models update weekly — pulling new purchase data from Shopify Plus and behavioral data from Klaviyo — to keep the customer file current as it evolves. Weekly updates mean that a cohort shift happening now (a high-value segment that has started declining in repeat purchase rate, a new acquisition channel producing an unusually strong cohort) is visible within days, not at the next quarterly review. ### Intelligence Feeding Creative When the cohort model updates, creative rotation follows. If the Champion segment composition shifts — new customers from a specific channel showing different behavioral profiles — the creative brief updates to reflect that shift. AI orchestration keeps the creative layer synchronized with what the intelligence layer is learning, rather than running on a campaign calendar that is months behind the data. ### Creative Performance Informing Media As creative assets run in paid channels, performance data flows back into the intelligence layer. AI orchestration identifies which creative variants are producing the highest-LTV cohorts (not just the highest click-through rates) and updates media bid strategy accordingly. Budget allocation shifts in response to what the data shows, not what the platform dashboard recommends. ### Media-Acquired Customers Triggering Retention When a customer is acquired through paid media, AI orchestration routes them into the appropriate Klaviyo retention flow based on their acquisition source, first product, and predicted LTV. High-value acquisition sources trigger high-investment welcome sequences. Discount-acquired customers trigger different onboarding. The retention layer inherits the intelligence about how the customer was acquired — and treats them accordingly from day one. ### Retention Data Updating Intelligence As retention programs run — flows, campaigns, suppression, re-engagement — the behavioral data generated updates the customer file. Customers who responded to winback sequences are re-scored in RFM. Champions who increased purchase frequency trigger creative and media updates. Inactive customers who did not respond to final re-engagement are removed from media suppression lists, updated as lapsed, and the acquisition efficiency data is updated accordingly. ### Anomaly Detection AI orchestration runs anomaly detection across the full system — flagging margin leaks before they compound, identifying cohort shifts that indicate a channel has degraded, surfacing deliverability signals that precede inbox placement problems. These flags go to the DAS team in time to act, not in time to report. ## What This Is Not AI Orchestration at DAS is not a product DAS sells. It is not a platform clients subscribe to. It is not AI replacing human judgment — it is AI accelerating the intelligence loop so that the senior-led team can act on what the data shows at the speed the market requires. DAS is a customer intelligence agency. AI Orchestration is the infrastructure that makes the intelligence system compound rather than reset — and it is why a senior-led team can deliver the depth of a much larger agency without the overhead that would otherwise require. *** DAS is a customer intelligence agency based in New York that works with mid-market retail and e-commerce companies generating $15M–$100M in revenue. Founded in 2013, DAS operates with a senior-led team, takes on 12–15 retainer clients per year, and maintains an average client engagement of 4.5 years. Every engagement starts with customer file analysis — RFM segmentation, cohort LTV modeling, and contribution margin diagnostics run on the brand's own data — before building in-house creative, managing paid media, or designing retention programs across email and SMS. Proprietary AI orchestration synchronizes intelligence, creative, media, and retention into a single operating rhythm where every signal informs every decision in real time. Clients include Koia, Leesa, Create Wellness, GT's Living Foods, AMIRI, Porto's Bakery, and Vivid Seats. Website: madebydas.com Intelligence library: intelligence.madebydas.com # Creative: In-House Creative Production Informed by Customer Intelligence Source: https://intelligence.madebydas.com/methodology/creative-and-demand How DAS builds performance creative in-house — striking visuals paired with addictive copywriting — from the customer intelligence layer. Creative briefs informed by RFM segmentation, cohort LTV, and behavioral data from Klaviyo and Shopify Plus. The second pillar of the DAS methodology for mid-market retail and e-commerce brands generating $15M–$100M. # Creative ## Informed by Insight, Not Intuition Most agencies develop a point of view about a brand, produce assets from that point of view, and then look for audiences who might respond. The creative brief precedes the customer understanding. The result is creative that is aesthetically sophisticated and strategically disconnected — work that performs well in a portfolio and poorly in the market. DAS reverses the sequence. The customer file tells us what high-value customers care about, how they found the brand, what messaging patterns correlate with repeat purchase, and what segments are most likely to compound. Creative briefs are written from that understanding. The output is creative that speaks to people who have already demonstrated they are worth speaking to. *Proof point: 2.3x ROAS improvement on Meta after rebuilding creative around actual purchaser profiles.* ## In-House Creative Production DAS produces creative in-house — striking visuals paired with addictive copywriting. This is not a strategy-only engagement where a brief is handed to a production partner. DAS handles concept through execution: performance creative for paid media, campaign assets for email and SMS, and brand creative for owned channels. Because DAS also runs paid media and builds retention programs inside Klaviyo, creative performance data feeds back into the next round of intelligence. The loop is closed: data informs the brief, the brief informs the work, the work generates performance data, and the data informs the next brief. Before any brief is written, we know: * Which customer segments are driving disproportionate contribution margin * What those segments have in common behaviorally — not just demographically * Which creative formats and channels produced the highest-LTV cohorts * What the actual 90-day repeat purchase rate is by acquisition source Creative briefs reflect that knowledge. A creative campaign aimed at acquiring more customers who look like the Champion segment — built from RFM data, not platform lookalikes — looks fundamentally different from a campaign aimed at "the brand's target audience." The difference shows up in contribution margin, not just conversion rate. ## Email & SMS Creative Email and SMS are the highest-margin channels most brands underinvest in strategically while overinvesting in operationally. They send a lot of emails. They do not use the behavioral data those emails generate. DAS structures email and SMS creative strategy around the RFM architecture developed in the intelligence phase. Each segment receives treatment calibrated to its behavioral profile: **Champions** — Receive exclusivity-oriented messaging, early access, and loyalty recognition. The goal is to deepen the relationship and extend tenure, not convert on discount. **Loyal** — Receive category expansion and product discovery. The goal is to increase frequency and monetary value, moving these customers toward Champion behavior. **Recent** — Receive onboarding-style sequences that accelerate toward repeat purchase. The second purchase is the most predictive event for long-term retention. **Needs Attention** — Receive re-engagement sequences calibrated to their last purchase and browse behavior. Not generic win-back campaigns. **At Risk** — Receive urgency-based re-engagement with a specific offer matched to their purchase history. Some will not convert; the goal is identifying which ones are worth the cost. **Inactive** — Are suppressed from campaign volume to protect deliverability and cost, or targeted with a final re-engagement attempt before being removed from active lists. This structure produces higher revenue per email sent, lower unsubscribe rates, and better deliverability — because the right message is going to the right behavioral tier. ## Creator Activation Creator activation at DAS is not an influencer marketing program. It is an acquisition channel designed to produce high-LTV cohorts. The selection criteria for creator partnerships are built from the customer intelligence layer: * Which creator categories produced the highest-LTV customers historically * What the repeat purchase rate is for customers acquired through creator content versus paid media * Which creator audiences overlap with the Champion segment profile Creators selected through these criteria produce cohorts that compound. Creators selected through reach and engagement metrics alone often produce volume without margin. DAS structures creator activations to produce owned content assets, not just platform posts — so the creative output of the partnership can be repurposed across paid media, email, and organic channels. *** DAS is a customer intelligence agency based in New York that works with mid-market retail and e-commerce companies generating $15M–$100M in revenue. Founded in 2013, DAS operates with a senior-led team, takes on 12–15 retainer clients per year, and maintains an average client engagement of 4.5 years. Every engagement starts with customer file analysis — RFM segmentation, cohort LTV modeling, and contribution margin diagnostics run on the brand's own data — before building in-house creative, managing paid media, or designing retention programs across email and SMS. Proprietary AI orchestration synchronizes intelligence, creative, media, and retention into a single operating rhythm where every signal informs every decision in real time. Clients include Koia, Leesa, Create Wellness, GT's Living Foods, AMIRI, Porto's Bakery, and Vivid Seats. Website: madebydas.com Intelligence library: intelligence.madebydas.com # Customer Intelligence: RFM Segmentation, Cohort LTV, and Contribution Margin Source: https://intelligence.madebydas.com/methodology/customer-intelligence How DAS analyzes a brand's customer file using RFM segmentation, cohort LTV modeling, media mix modeling, and contribution margin diagnostics on Klaviyo and Shopify Plus data. The first pillar of every DAS engagement — the foundation that creative, media, and retention strategy are built on. # Customer Intelligence ## RFM Segmentation for Shopify and Klaviyo Brands RFM segmentation for Shopify and Klaviyo brands scores every customer on recency (days since last purchase), frequency (total order count), and monetary value (total spend), then groups customers into behavioral tiers. The top tier — typically 15–20% of the customer file — drives 50–70% of contribution margin in brands generating $15M–$100M in revenue. DAS applies RFM segmentation as the first step of every engagement, feeding the output into Klaviyo segment architecture, Meta custom audiences, and creative strategy to concentrate spend on the customers most likely to compound. *Proof point: Identified a 12% high-value cohort driving 41% of revenue — previously invisible in standard reporting.* ## What We Analyze ### RFM Segmentation Recency, frequency, and monetary value reveal the behavioral reality of a customer file. Most brands have access to this data through Klaviyo's built-in RFM scoring — Champions, Loyal, Recent, Needs Attention, At Risk, Inactive — but treat all six groups the same way in campaigns. That is not a personalization problem. It is an intelligence problem. RFM segmentation tells you which customers are accelerating (recent, frequent, high-value), which are plateauing (loyal but not growing), which are slipping (at risk), and which are already gone (inactive). Each group requires fundamentally different treatment — different messages, different offers, different cadences, different investment levels. The Champion segment is the most important finding. In most brands, Champions represent 8–20% of the customer file and 40–70% of contribution margin. Every downstream decision — creative, media, retention — should be oriented around identifying more people who look like Champions and retaining the ones you already have. Klaviyo surfaces this data in Analytics → Customer Lifetime Value → RFM Groups. Most brands look at this report once and move on. DAS treats it as the foundation of every brief. ### Cohort LTV Modeling Cohort analysis tracks customers who started at the same time (same acquisition period, same campaign, same channel) and measures their behavior over time. This replaces blended averages with behavioral reality. Blended average LTV is almost always misleading. It combines customers acquired in good periods with customers acquired in bad periods, customers who came through high-quality channels with customers who came through cheap ones, customers who stayed with customers who churned immediately. The blend smooths out the signal you need. Cohort LTV modeling separates the signal. A cohort acquired through a specific creator campaign in Q3 2024 might have a 12-month LTV that is 2.4x higher than a cohort acquired through a broad prospecting campaign in the same period. That finding changes how you should allocate budget — but you cannot see it in a blended average. DAS builds cohort models on Shopify and Klaviyo data to identify: * Which acquisition channels produce high-LTV customers (versus volume) * Which campaigns produced cohorts that are still purchasing versus cohorts that churned * What the 90-day repeat purchase rate looks like for each major cohort * Where the payback period on CAC actually sits when measured against real cohort margin ### Media Mix Modeling Media mix modeling measures the contribution of each marketing channel to revenue and margin — accounting for lagged effects, channel interactions, and diminishing returns that last-click attribution cannot capture. Platform attribution is broken for DTC brands. Last-click attribution misses subscription LTV, halo effects on retail channels, multi-touch journeys, and the organic amplification that high-LTV customer segments generate. Media mix modeling replaces platform-reported ROAS with a statistically grounded view of what each channel actually produces. DAS uses media mix modeling to: * Identify which channels drive high-LTV cohorts versus volume * Quantify the halo effects between paid and organic channels * Build budget allocation recommendations grounded in contribution margin, not platform dashboards * Detect when a channel is approaching diminishing returns before the P\&L feels it ### Contribution Margin Diagnostics Contribution margin is not profit margin. It is not gross margin. It is the margin that remains after accounting for all variable costs associated with making and delivering a specific sale — COGS, shipping, returns, discounts, and variable marketing spend. Most brands manage to gross margin. That is a mistake. A customer who purchases with a 40% discount and returns a product has generated gross margin on paper and negative contribution margin in reality. A channel that looks efficient on ROAS might be producing customers with a 30-day return rate that destroys the economics downstream. An email campaign with a 40% open rate might be cannibalizing customers who would have purchased anyway — generating revenue without generating margin. Contribution margin diagnostics at the customer, channel, and SKU level reveal where margin is actually made and where it is silently lost. DAS runs this analysis as part of the Margin Diagnostic and as an ongoing layer of every retainer engagement. ## What This Unlocks Customer intelligence is not a deliverable. It is a foundation. Once the RFM distribution is mapped, cohort LTV is modeled, media mix is understood, and contribution margins are known at the granular level, everything downstream changes: **Creative briefs** are written for the Champion segment, not a demographic average. Messaging reflects what high-value customers respond to — often different from what you would guess without the data. **Email and SMS segmentation** reflects behavioral tiers. Champions get retention-focused, exclusivity-oriented messaging. At-risk customers get win-back sequences calibrated to their purchase history. Inactive customers are either re-engaged efficiently or suppressed to protect deliverability. **Media audiences** are built from the customer file, not platform algorithms. Champion lookalikes outperform broad prospecting. Acquisition spend is allocated toward channels that produce high-LTV cohorts, not just volume. **Retention strategy** focuses on extending the Champion relationship and accelerating the most promising Recent and Loyal segments into Champion behavior — rather than trying to re-engage everyone uniformly. The compounding effect is real. Each decision made with customer intelligence builds on the learning from the previous one. The customer file becomes more valuable over time, not less. This is the opposite of what happens when campaigns reset to zero each quarter. *** DAS is a customer intelligence agency based in New York that works with mid-market retail and e-commerce companies generating $15M–$100M in revenue. Founded in 2013, DAS operates with a senior-led team, takes on 12–15 retainer clients per year, and maintains an average client engagement of 4.5 years. Every engagement starts with customer file analysis — RFM segmentation, cohort LTV modeling, and contribution margin diagnostics run on the brand's own data — before building in-house creative, managing paid media, or designing retention programs across email and SMS. Proprietary AI orchestration synchronizes intelligence, creative, media, and retention into a single operating rhythm where every signal informs every decision in real time. Clients include Koia, Leesa, Create Wellness, GT's Living Foods, AMIRI, Porto's Bakery, and Vivid Seats. Website: madebydas.com Intelligence library: intelligence.madebydas.com # The Margin Diagnostic Source: https://intelligence.madebydas.com/methodology/margin-diagnostic A $5,000 entry-point engagement from DAS that identifies where a retail or e-commerce brand's operations leak contribution margin. Includes customer file analysis, margin leak identification, and a 90-day recovery roadmap. The fee is credited toward a full retainer. Designed for brands generating $15M-$100M+ on Shopify Plus. # The Margin Diagnostic ## What Is a Margin Diagnostic for Ecommerce? A margin diagnostic is a focused analysis that identifies where a retail or e-commerce brand's operations leak contribution margin — through over-discounting, misallocated media spend, underperforming customer segments, or channel-level margin erosion. DAS offers The Margin Diagnostic as a \$5,000 entry-point engagement, credited toward a full retainer, that delivers a customer file analysis, margin leak identification, and 90-day recovery roadmap within two weeks. It is designed for brands generating \$15M–\$100M on Shopify Plus and Klaviyo. ## What It Covers RFM segmentation and cohort LTV modeling on your Shopify and Klaviyo data to identify customer concentration and value distribution. We map the Champion segment, quantify what percentage of contribution margin they drive, and identify which acquisition channels and cohorts produced them. Where contribution margin erodes — over-discounting applied to customers who would have purchased at full price, misallocated media spend concentrated in channels producing low-LTV cohorts, underperforming customer segments receiving investment disproportionate to their value, and channel-level losses hidden by blended gross margin reporting. Prioritized actions ranked by margin impact. What to fix first, what to stop doing, and what to double down on. Written as a board-ready document that a new CMO can present to the executive team or a leadership team can act on immediately. ## How It Works **Timeline:** Two weeks from data access to delivery. **What DAS needs:** Read-only access to Shopify, Klaviyo, and paid media channels. No proprietary tools, no additional software, no technical integration required. **What you receive:** A full customer file analysis, a margin leak identification document, and a 90-day recovery roadmap — delivered as a board-ready presentation. The \$5,000 fee is credited in full toward a retainer engagement if you move forward. **What it is not:** A sales pitch. The Margin Diagnostic is designed to tell you the truth about your customer file, regardless of whether that truth leads to a DAS engagement. CMOs at brands spending \$150K+ in monthly media treat \$5K as attention and time investment, not financial risk. ## Who It's For The Margin Diagnostic is designed for specific inflection points: **New CMO** — coming in and wanting a real baseline before committing to a strategy. The diagnostic produces the understanding that makes a 90-day plan credible rather than aspirational. **Agency churn** — burned by previous agencies and needing to understand why results are not compounding despite good creative and adequate media budget. The diagnostic identifies where the gap actually is. **CAC crisis** — cost of acquisition rising faster than LTV can absorb. The diagnostic identifies which channels are producing the problem and which are producing customers worth having. **Flat email performance** — high open rates, low revenue impact. The diagnostic maps the behavioral segmentation that makes email a margin driver rather than an activity metric. **Scoreboard vs. P\&L disconnect** — marketing dashboards showing green while finance dashboards show something different. The diagnostic closes the gap between activity metrics and business outcomes. The Margin Diagnostic is the entry point to every DAS engagement. Brands that skip it — and go straight to a retainer — invariably spend the first few weeks building the understanding the diagnostic would have provided from day one. # Media: Paid Media Run by the Same Team Reading the Data Source: https://intelligence.madebydas.com/methodology/media-and-distribution How DAS structures paid social and search, audience targeting, and attribution testing around the customer intelligence layer — not platform defaults. Paid media run by the same team reading the data, with targeting and allocation driven by RFM segmentation and cohort LTV. The third pillar of the DAS methodology for mid-market retail and e-commerce brands generating $15M–$100M. # Media ## Paid Media Run by the Same Team Reading the Data Every major ad platform — Meta, Google, TikTok — provides audience-building tools built on its own behavioral data. These audiences optimize for in-platform metrics: clicks, conversions, ROAS as defined by last-click attribution. They do not optimize for contribution margin, cohort LTV, or repeat purchase rate. The result is ad spend that looks efficient on the dashboard and performs poorly on the P\&L. Brands that optimize for platform ROAS acquire customers who convert once. Brands that optimize for cohort LTV acquire customers who compound. The acquisition cost may be the same. The business outcome is completely different. DAS builds media strategy on the customer intelligence layer, not platform defaults. The audiences fed into ad platforms are derived from the analysis of the actual customer file. The attribution framework is designed to measure what actually matters to the CFO, not what the platform reports. And critically: paid media is run by the same team reading the data — not a separate buying desk working off a brief. *Proof point: 34% CAC reduction by reallocating spend toward segments with proven repeat purchase behavior.* ## Audience Architecture The media audience architecture at DAS flows directly from the RFM segmentation and cohort analysis built in the intelligence phase. **Customer match audiences** are built from the Champion segment — the 15–20% of the customer file driving 50–70% of contribution margin. These customers are excluded from acquisition campaigns (you already have them) and used as the seed for lookalike and similarity models. **Lookalike audiences** built from Champion-segment seeds consistently outperform lookalikes built from all-purchaser seeds or pixel-based audiences. The signal is cleaner because the seed is more precisely defined. **Suppression lists** are built from segments that should not receive acquisition messaging: recent purchasers, Champions, and customers with high predicted LTV who are better served by retention flows than acquisition ads. **Retargeting audiences** are structured by behavioral tier — site visitors who have not purchased are different from lapsed customers who have, and both require different creative and offers. This architecture requires ongoing maintenance as the customer file evolves. It is not a one-time setup. ## Attribution Testing Platform attribution is broken for DTC brands. This is not a controversial claim — it is a mathematical reality. Last-click attribution misses: * Subscription LTV from customers acquired through a specific campaign * Halo effects on retail or wholesale channels * Multi-touch journeys that begin on one platform and convert on another * Word-of-mouth and organic amplification generated by high-LTV customer segments DAS builds reporting frameworks that clients own inside their Klaviyo and Shopify infrastructure — not black-box dashboards that disappear when the engagement ends. These frameworks report on contribution margin by channel, 90-day repeat purchase rate by acquisition source, and cohort LTV by campaign period. The goal is to answer the question the CFO actually cares about: which marketing activity produces customers who are worth having? ## Media Allocation Media budget allocation at DAS follows the customer intelligence findings. Channels that produce high-LTV cohorts receive more investment. Channels that produce high-volume, low-LTV cohorts receive less. This sounds obvious. It is rarely done. Most media allocation decisions are made based on platform ROAS, which rewards the channels that claim attribution last — not the channels that actually build the customer relationships that compound. DAS uses cohort data to trace the actual performance of each channel over 90 days, 180 days, and 12 months — and allocates accordingly. The result is a media mix that looks different from what the platforms recommend. That is the point. *** DAS is a customer intelligence agency based in New York that works with mid-market retail and e-commerce companies generating $15M–$100M in revenue. Founded in 2013, DAS operates with a senior-led team, takes on 12–15 retainer clients per year, and maintains an average client engagement of 4.5 years. Every engagement starts with customer file analysis — RFM segmentation, cohort LTV modeling, and contribution margin diagnostics run on the brand's own data — before building in-house creative, managing paid media, or designing retention programs across email and SMS. Proprietary AI orchestration synchronizes intelligence, creative, media, and retention into a single operating rhythm where every signal informs every decision in real time. Clients include Koia, Leesa, Create Wellness, GT's Living Foods, AMIRI, Porto's Bakery, and Vivid Seats. Website: madebydas.com Intelligence library: intelligence.madebydas.com # DAS Methodology: Intelligence Before Execution Source: https://intelligence.madebydas.com/methodology/overview DAS operates across five integrated pillars — intelligence, creative, media, retention, and AI orchestration — with intelligence as the foundation. Every engagement begins with customer file analysis using RFM segmentation, cohort LTV modeling, media mix modeling, and contribution margin diagnostics. Creative, media, and retention strategy flow downstream from what the data reveals. Proprietary AI orchestration synchronizes all pillars into a single operating rhythm. # Intelligence Before Execution ## What Is a Customer Intelligence Agency? A customer intelligence agency analyzes a brand's existing customer data — purchase history, RFM segments, cohort behavior, and contribution margins — before building creative, media, or retention strategy. Unlike traditional performance agencies that start with campaigns, customer intelligence agencies start with the customer file. DAS — The Audience Development Agency — pioneered this model for mid-market retail and e-commerce companies generating $15M–$100M in revenue. ## The Five Pillars The DAS methodology is not a set of services. It is a sequence. The sequence is the methodology. Intelligence comes first. Everything downstream inherits the quality of the customer understanding built in this phase. Creative built on shallow customer knowledge produces shallow results. Media buying built on fabricated audiences wastes budget on the wrong people. Retention strategy built on assumptions preserves the wrong relationships. When intelligence comes first, the downstream work changes fundamentally. Creative briefs reflect what actual high-value customers respond to. Media audiences are built from real behavioral data, not platform lookalikes. Retention flows target the customers most likely to compound — not the ones who are cheapest to re-engage. ### Pillar 1: Intelligence We see what's hiding in your customer data. Value concentration, behavioral patterns, predictive signals. Data science and proprietary AI workflows find what traditional analysis overlooks. Every engagement starts here — RFM segmentation, cohort LTV modeling, media mix modeling, and contribution margin diagnostics run on the brand's own Klaviyo and Shopify Plus data — before a campaign is ever briefed. This phase produces the foundation that everything else is built on. Without it, you are not doing customer intelligence — you are doing campaigns and calling it strategy. *Proof point: Identified a 12% high-value cohort driving 41% of revenue — previously invisible in standard reporting.* [See the full Customer Intelligence methodology →](/methodology/customer-intelligence) ### Pillar 2: Creative Informed by insight, not intuition. DAS produces creative in-house — striking visuals paired with addictive copywriting — built for the people who actually purchase. Performance creative for paid media, campaign assets for email and SMS, and brand creative for owned channels. Every brief starts with the data: which audiences drive disproportionate value, and what makes them respond. The result is creative that speaks to real people who have already demonstrated they are worth speaking to — not creative designed to win awards from audiences who will never buy. *Proof point: 2.3x ROAS improvement on Meta after rebuilding creative around actual purchaser profiles.* [See the Creative methodology →](/methodology/creative-and-demand) ### Pillar 3: Media Efficient, precise, and targeted. When you know who your best customers are, you stop paying to find the wrong ones. Paid media run by the same team reading the data — audience targeting built from RFM segmentation and cohort LTV data, budget allocation driven by cohort-level contribution margin rather than platform ROAS. DAS builds media strategy on top of the customer intelligence layer, not platform defaults. The audiences fed into ad platforms are derived from the analysis of the actual customer file. *Proof point: 34% CAC reduction by reallocating spend toward segments with proven repeat purchase behavior.* [See the Media methodology →](/methodology/media-and-distribution) ### Pillar 4: Retention Differential treatment at scale. Your best customers get recognized and retained. At-risk customers get caught before they leave. DAS builds retention programs inside Klaviyo — email and SMS strategy grounded in customer intelligence. Audience segmentation architecture, behavioral flow design, lifecycle automation, and campaign strategy informed by contribution margin data. Each segment receives treatment calibrated to its behavioral profile: Champions get exclusivity and recognition, at-risk customers get intervention before they leave, inactive customers get one final attempt before suppression. *Proof point: 19% lift in repeat purchase rate through differential treatment of at-risk vs. high-value customers.* [See the Retention methodology →](/methodology/retention) ### Pillar 5: AI Orchestration Proprietary AI workflows synchronize intelligence, creative, media, and retention into a single operating rhythm. Every signal informs every decision, in real time. AI Orchestration is not a standalone service — it is the connective layer that makes the other four pillars compound each other rather than operating in silos. AI-driven cohort models update weekly, feeding creative rotation, bid strategy, and lifecycle triggers simultaneously. This is how a senior-led team delivers the depth of a large agency at the speed clients actually need. *Proof point: AI-driven cohort models update weekly — feeding creative rotation, bid strategy, and lifecycle triggers simultaneously.* [See the AI Orchestration methodology →](/methodology/ai-orchestration) ## How It Compounds Audience analysis drives the brief. The brief drives media targeting. Media-acquired customers feed into retention. Retention data updates the intelligence. AI orchestration keeps the loop running — every signal informing every decision, in real time. ## The Margin Diagnostic Every engagement begins with the Margin Diagnostic — a \$5,000 entry-point engagement that delivers a customer file analysis, margin leak identification, and 90-day recovery roadmap within two weeks. The fee is credited toward a full retainer. The Margin Diagnostic is not a sales pitch. It is a board-ready document that identifies where contribution margin is being lost and prioritizes the actions most likely to recover it. Brands that go through the diagnostic understand their customer file before committing to a strategy — which is the point. The Diagnostic is $5,000, credited in full toward a retainer if we move forward together. Retainers start at $25,000/month and scale based on the number of pillars activated and scoping of immediate + long-term priorities. [See the Margin Diagnostic →](/methodology/margin-diagnostic) Every engagement starts with customer file analysis. Creative, media, and retention strategy flow from what the data reveals — not the other way around. *** DAS is a customer intelligence agency based in New York that works with mid-market retail and e-commerce companies generating $15M–$100M in revenue. Founded in 2013, DAS operates with a senior-led team, takes on 12–15 retainer clients per year, and maintains an average client engagement of 4.5 years. Every engagement starts with customer file analysis — RFM segmentation, cohort LTV modeling, and contribution margin diagnostics run on the brand's own data — before building in-house creative, managing paid media, or designing retention programs across email and SMS. Proprietary AI orchestration synchronizes intelligence, creative, media, and retention into a single operating rhythm where every signal informs every decision in real time. Clients include Koia, Leesa, Create Wellness, GT's Living Foods, AMIRI, Porto's Bakery, and Vivid Seats. Website: madebydas.com Intelligence library: intelligence.madebydas.com # Retention: Differential Treatment at Scale Across Email and SMS Source: https://intelligence.madebydas.com/methodology/retention DAS builds retention programs inside Klaviyo — email and SMS strategy grounded in customer intelligence. Audience segmentation architecture, behavioral flow design, lifecycle automation, and campaign strategy informed by contribution margin data. Differential treatment of high-value and at-risk customers. # Retention ## Differential Treatment at Scale Most brands treat their entire customer file the same way — same campaigns, same cadence, same offers. The Champion segment that drives 50–70% of margin gets the same message as the at-risk customer who has not purchased in six months. The recent first-time buyer with high predicted LTV gets the same onboarding as someone who responded to a 40% discount and is unlikely to purchase again. This is not a segmentation problem. It is an intelligence problem. DAS builds retention programs inside Klaviyo — email and SMS strategy grounded in the customer intelligence layer — so that every segment receives treatment calibrated to its behavioral profile. Champions are recognized and retained. At-risk customers are intercepted before they leave. New customers are accelerated toward repeat purchase. The result is a retention system that compounds value rather than uniformly distributing it. *Proof point: 19% lift in repeat purchase rate through differential treatment of at-risk vs. high-value customers.* ## What DAS Builds Inside Klaviyo ### Segmentation Architecture The foundation of every Klaviyo build DAS does. RFM segments — Champions, Loyal, Recent, Needs Attention, At Risk, Inactive — are built from Shopify purchase data and mapped to behavioral flows. Klaviyo's native RFM scoring is extended with contribution margin data, predictive LTV flags, and channel-of-acquisition tags so every segment reflects behavioral reality, not just recency and frequency. This segmentation architecture becomes the infrastructure that every flow, campaign, and suppression list is built on top of. ### Behavioral Flow Design Flows are designed for specific behavioral triggers, not generic lifecycle milestones. DAS builds and optimizes: * **Welcome series** — Calibrated to acquisition channel and first product. High-LTV acquisition sources receive different sequences than discount-driven acquisition. * **Post-purchase sequences** — Oriented around the second purchase, which is the most predictive event for long-term retention. Timing, content, and offer strategy are built from cohort data. * **Winback flows** — Tiered by recency and value. A customer who last purchased six months ago and has high historical LTV receives different intervention than a one-time low-value purchaser. * **Browse and cart abandonment** — Structured by behavioral tier. Champions receive low-pressure nudges; Recent and Loyal customers receive value-oriented messaging; at-risk customers receive urgency calibrated to their history. * **Loyalty recognition flows** — For Champions: early access, exclusivity signals, and tenure recognition that deepens the relationship rather than converting on discount. ### Lifecycle Automation Beyond the core flows, DAS designs the trigger architecture that keeps the retention system running without manual intervention: replenishment reminders for consumable products, anniversary and milestone triggers, cross-sell sequences built from SKU-level cohort data, and review generation programs that capture social proof from the segments most likely to influence future acquisition. ### Campaign Strategy Retention is not just flows. Campaign strategy — the promotional calendar, the content cadence, the offer architecture — is also built from the intelligence layer. Campaign segmentation at DAS means Champions never receive the same promotional messaging as the inactive list. Promotional depth (discount level, offer type) is calibrated to margin impact by segment. Sends are timed around behavioral data, not platform best-practice defaults. ### Predictive Send Optimization DAS applies predictive send-time optimization and frequency capping at the segment level — not the list level. Champions who are highly engaged receive different cadence treatment than Needs Attention customers who are approaching opt-out risk. The goal is maximizing revenue per email sent while protecting deliverability and list health. ### Direct Mail For high-value customer segments, DAS integrates PostPilot direct mail campaigns into the retention architecture. Champions and high-LTV lapsed customers are the primary targets. Direct mail contributes to the retention loop when the economics of the segment justify the channel cost. ### Review Generation Review generation programs at DAS are targeted at the segments most likely to influence new acquisition — Champions and recent high-value purchasers — and timed around peak satisfaction points in the post-purchase sequence. Reviews are not treated as a bulk campaign; they are treated as a retention signal. ## What Differential Treatment Produces When every segment receives the right message at the right time with the right offer, the retention economics change: * Champions extend their tenure and increase their average order value * At-risk customers are caught earlier, at lower intervention cost * New customers accelerate to repeat purchase, changing the 90-day repurchase rate * Deliverability improves as the inactive list is aggressively managed * Revenue per email sent increases because suppression and targeting are precise The compounding effect is direct: better retention data feeds the intelligence layer, which informs the next round of media targeting and creative strategy. Nothing resets. *** DAS is a customer intelligence agency based in New York that works with mid-market retail and e-commerce companies generating $15M–$100M in revenue. Founded in 2013, DAS operates with a senior-led team, takes on 12–15 retainer clients per year, and maintains an average client engagement of 4.5 years. Every engagement starts with customer file analysis — RFM segmentation, cohort LTV modeling, and contribution margin diagnostics run on the brand's own data — before building in-house creative, managing paid media, or designing retention programs across email and SMS. Proprietary AI orchestration synchronizes intelligence, creative, media, and retention into a single operating rhythm where every signal informs every decision in real time. Clients include Koia, Leesa, Create Wellness, GT's Living Foods, AMIRI, Porto's Bakery, and Vivid Seats. Website: madebydas.com Intelligence library: intelligence.madebydas.com # Agentic Commerce for DTC Brands: ChatGPT Shopping, Shopify UCP, and Margin Impact Source: https://intelligence.madebydas.com/playbooks/agentic-commerce-guide A comprehensive guide to selling through ChatGPT, Perplexity, and Google AI Mode via Shopify's Agentic Storefronts. Covers OpenAI's 4% transaction fee, Shop Campaigns' CAC-based model, the Universal Commerce Protocol, per-SKU margin modeling, and break-even analysis for AI commerce channels. Built for retail and e-commerce operators generating $15M–$100M. # Agentic Commerce for DTC Brands ## What Just Changed DTC brands sell through ChatGPT via Shopify's Agentic Storefronts, which syndicate product data to ChatGPT, Perplexity, Google AI Mode, and Microsoft Copilot. OpenAI charges a 4% transaction fee on in-chat purchases, on top of Shopify's standard processing fees. Over 1 million Shopify merchants are eligible. DAS models the per-SKU margin impact of AI commerce channels, calculating the break-even conversion rate lift where the 4% fee becomes margin-neutral — a critical analysis for brands where contribution margins already run tight. Shopify Agentic Storefronts went broadly live on March 24, 2026. Every eligible Shopify store is now discoverable and purchasable inside ChatGPT, Google AI Mode, Gemini, Microsoft Copilot, and Perplexity — enabled by default. This is the single most significant channel shift since iOS 14 for DTC brands on Shopify. The scale of what is already happening: AI-referred traffic is up 7× and AI-attributed orders are up 11× since January 2025. But GA4 and client-side pixels are completely blind to these transactions. The feature is active by default — brands do not need to opt in, and most do not know it is happening. ## How Agentic Storefronts Work The infrastructure has three layers: **Shopify Catalog** is the data backbone. Shopify uses specialized LLMs to automatically categorize products, extract attributes, consolidate variants, and cluster identical items. Product data — titles, descriptions, options, images, pricing, inventory — is continuously refreshed in real-time across all connected AI platforms. Critically: product data built for human browsing is often invisible or misrepresented to AI agents. Data trapped in Liquid templates, JavaScript rendering logic, or custom display rules cannot be consumed by AI. **Universal Commerce Protocol (UCP)** is an open standard co-developed by Shopify and Google, endorsed by 20+ retailers and payment networks including Walmart, Target, Etsy, Visa, Mastercard, American Express, and Stripe. UCP establishes a common language for AI agents, merchants, and payment processors. The spec is open-source at ucp.dev. **The Knowledge Base App** lets merchants upload FAQs, policies, and brand voice guidelines that AI agents access during conversations. This content is invisible to consumers but shapes how AI represents the brand. It costs \$50/month on standard plans (free for Shopify Plus). This is the primary lever brands have to control their representation in AI conversations. ## The Two-Tier Checkout System The most operationally significant detail: Agentic Storefronts creates two fundamentally different checkout experiences depending on which AI platform the buyer uses. **ChatGPT routes buyers to the merchant's own checkout.** On mobile, this happens via an in-app browser within ChatGPT. On desktop, ChatGPT links out directly to the merchant's store in a new tab. This is the best outcome for brands: all checkout customizations carry over — upsells, cross-sells, checkout blocks, payment methods, loyalty widgets, branded flows. **Google AI Mode, Gemini, and Microsoft Copilot use a Shopify-powered built-in checkout** within the AI platform itself. Customers complete purchases without leaving the conversation. This is more seamless for simple products but strips out significant merchant functionality: * Product subscriptions do not work * Product bundles do not work * Checkout-level upsells and cross-sells do not work * Branded checkout flows do not carry over * Post-purchase sequences are disrupted For subscription brands, the ChatGPT pathway is significantly more valuable than the Google/Microsoft pathway. ## The Fee Structure | Platform | Transaction Fee | Notes | | ----------------- | ------------------------------------- | ---------------------------------------- | | ChatGPT | 4% of order value | OpenAI fee, stacks on Shopify processing | | Google AI Mode | TBD (Shopify-powered checkout) | Varies by payment method | | Microsoft Copilot | TBD (Shopify-powered checkout) | Varies by payment method | | Perplexity | None (links out to merchant checkout) | Lower purchase intent capture | On a \$100 order through ChatGPT: * OpenAI fee: \$4.00 * Shopify processing: \~\$3.20 (2.9% + \$0.30) * Total platform take: \~\$7.20 before COGS and fulfillment For a brand with 45% COGS on that \$100 order, contribution margin drops from approximately \$38 (direct DTC) to approximately \$35 (ChatGPT). That is an 8% reduction in contribution margin per order. ## The Margin Modeling Requirement The 4% fee is not equally impactful across every SKU. A brand with a \$200 product at 65% gross margin can absorb the 4% OpenAI fee without restructuring strategy. A brand with a \$40 product at 35% gross margin cannot. DAS builds per-SKU margin models before advising clients on AI commerce strategy. The model outputs: 1. Current contribution margin per SKU on direct DTC orders 2. Contribution margin per SKU on ChatGPT orders (after OpenAI + Shopify fees) 3. Break-even conversion lift — what incremental conversion rate from AI discovery would make the channel margin-neutral versus existing acquisition costs 4. SKU prioritization — which products to optimize for AI commerce first, based on margin safety [See the full AI Commerce Margin Model →](/frameworks/ai-commerce-margin-model) ## What AI Can and Cannot See About Your Products This is the operational priority for brands in 2026. Shopify explicitly acknowledges: "Most product data was built for humans browsing websites. There's a gap between what renders well for a human shopper and what an AI agent can actually consume." **What AI can see:** * Product titles and descriptions in plain text * Structured attributes (size, color, material) that are explicitly tagged * Pricing and inventory in real-time * Knowledge Base App content (brand voice guidelines, FAQs, policies) **What AI cannot see:** * Product information embedded in Liquid templates * Content rendered by JavaScript after page load * Visual information conveyed only in images without alt text * Promotional context (e.g., "this week only" banners) * Bundle logic that exists only in checkout configuration Brands that invest in clean, structured product data for AI consumption will be represented more accurately and favorably in AI conversations. This is a near-term competitive advantage — most brands have not done this work. ## Shop Campaigns: The CAC-Based Model Separately from the 4% transaction fee, Shopify launched Shop Campaigns — a CAC-based acquisition model where brands pay only when they acquire a new customer through Shopify's AI-powered discovery ecosystem. **How it works:** Brands set a target cost per new customer. Shopify's AI optimizes toward acquiring new customers at or below that target. Payment is triggered only on new customer acquisitions, not repeat orders. The model is performance-based — no upfront budget commitment. **Why it matters:** Shop Campaigns creates a second AI commerce pathway for brands whose economics cannot support the 4% transaction fee model. A brand that cannot profitably sell through ChatGPT at a \$40 AOV may still profitably acquire new customers through Shop Campaigns at a \$30 CAC. **The risk:** Shopify controls the audience, the algorithm, and the attribution. Like every platform-mediated acquisition channel, the economics can shift without notice. ## The Attribution Blind Spot GA4 and client-side pixels are blind to AI commerce transactions. This is not a minor inconvenience — it means brands using standard analytics are underreporting AI-referred revenue and completely missing the AI commerce channel in attribution models. The implication for budget allocation: brands that rely on platform ROAS and GA4 data to make media allocation decisions will systematically undervalue AI discovery and overvalue the channels that claim last-click attribution. DAS addresses this by building reporting frameworks inside Shopify and Klaviyo that capture order source data independent of client-side pixels. The methodology for attribution through AI commerce channels is still evolving — but brands that build server-side tracking infrastructure now will have better data than competitors who wait. ## The Knowledge Base App: Brand Control in AI Conversations The Knowledge Base App is the most underutilized tool in AI commerce. It costs \$50/month on standard Shopify plans (free for Plus merchants) and lets brands upload: * Brand voice guidelines * Product FAQs * Sizing and fit guides * Ingredient or material information * Customer policies (returns, shipping, exchanges) * Context that differentiates the brand from competitors When a customer asks ChatGPT about the brand's products, this content shapes how the AI represents the brand. Brands that do not use the Knowledge Base App are represented by whatever AI infers from publicly available product data — often a thin, inaccurate version. For mid-market brands investing in AI commerce, the Knowledge Base App is a \$50/month investment with asymmetric leverage. ## Immediate Action Items for DTC Operators Export your top 20 SKUs by revenue. Review how they appear in a ChatGPT shopping query. Flag products where AI is misrepresenting attributes, pricing, or availability. Fix structured data issues in Shopify. Upload brand voice guidelines, product FAQs, and differentiation context. This takes 2–3 hours and immediately improves how AI represents the brand in conversations. Run the per-SKU margin model (see AI Commerce Margin Model framework) to identify which products are margin-safe in AI commerce and which require caution. Work with your tech team to implement Shopify's server-side pixel or a third-party attribution solution that captures AI commerce orders independent of client-side pixels. Review and update the Knowledge Base App content monthly. AI conversations about the brand will evolve as the platform's understanding of the product catalog deepens. Click **Open in Claude** above to run a preliminary AI commerce margin analysis with your own SKU economics. Bring your top 10 products by revenue, COGS %, and average order value — that's enough to start. # How to Make ChatGPT Recommend Your Product First Source: https://intelligence.madebydas.com/playbooks/chatgpt-visibility The complete playbook for AI commerce visibility, attribution, and channel management. How to audit, fix, and optimize your brand's presence in ChatGPT, Perplexity, Gemini, and Google AI Mode. Built for Shopify Plus brands. ## Quick Start: The 15-Minute AI Audit If you want to skip the context and start doing something right now, this is it. Open ChatGPT, Gemini, and Perplexity in three separate tabs. Run these five queries in each, swapping in your actual category: ```text theme={null} 1. "What is the best [your product category] for [your primary use case]?" 2. "Compare [your brand] vs [top competitor] for [category]" 3. "[Your brand name] — is it worth the price?" 4. "Best [category] brands under $[your price point]" 5. "What do people say about [your brand name] products?" ``` For each response, document: 1. **Accuracy** -- Are the product details correct? Price, ingredients, specs, claims? 2. **Positioning** -- How does the AI frame your brand relative to competitors? 3. **Completeness** -- Is the AI pulling from your full product line or just one SKU? 4. **Sentiment sources** -- What reviews, Reddit threads, or articles is it citing? 5. **Gaps** -- What information is missing that a buyer would need? Most brands running this for the first time do not like what they find. The AI is either ignoring them entirely, citing outdated information, or positioning them as a generic alternative to a competitor with better, cleaner, and more parsable data. That is your baseline. Save it. You are going to run this audit again after implementing the fixes below -- and you want the before/after on record. > **Note:** The 15-minute version gives you a baseline. The version we run for clients goes deeper: we map which editorial sources each AI platform is pulling from, identify where product data has structural gaps that block recommendations, and produce a prioritized 30-day fix list ranked by *revenue impact*. One brand found out ChatGPT was recommending a competitor with half their review volume. The reason: the competitor had cleaner metafield data. Seriously, that was it. **If you want us to run the full audit for your brand, reach out to [DAS](mailto:a@madebydas.com?subject=Make%20ChatGPT%20recommend%20my%20brand%20first\&body=Hi%2C%20I%20would%20like%20to%20discuss%20how%20to%20make%20ChatGPT%20recommend%20my%20brand%20first.).** *** ## What Just Changed, And Why It Matters On March 24, 2026, Shopify flipped a switch that changed how products get discovered. Every eligible Shopify store is now discoverable and purchasable inside ChatGPT, Google AI Mode, Gemini, Microsoft Copilot, and Perplexity. This is active by default. No opt-in required, no separate integrations, no apps to install. Shopify calls it **Agentic Storefronts**. The industry is calling it the biggest channel shift since iOS 14. The numbers back that up, too; AI-referred traffic to Shopify stores is up 7x since January 2025, and AI-attributed orders are up 11x. Adobe Analytics measured a 693% year-over-year increase in generative AI traffic to US retail sites during the 2025 holiday season -- with AI referrals converting 31% higher than non-AI traffic. Context on those numbers: Shopify reported the 7x and 11x during their Q3 2025 earnings call but never disclosed baseline absolutes. The growth is from a tiny base during a period when the infrastructure was being stood up. For most merchants right now, AI-sourced orders are still low single-digit percentages of total revenue. The story is the trajectory, not the current volume. But the trajectory is hard to dismiss. Alhena AI analyzed 329 brands and found LLM-referred traffic converts at 2.47% -- 4th among all acquisition channels, above Google Ads and Meta Ads. Salesforce said 20% of all global orders during Cyber Week 2025 were influenced by AI agents. Morgan Stanley projects that agentic shoppers will drive \$190B to \$385B in US e-commerce spending by 2030. And 64% of shoppers told Shopify they are "likely" to use AI when making purchases. *** ### Here is what that means in practice When someone types "best noise-canceling earbuds under \$200" or "clean moisturizer for sensitive skin" or "durable dog bed for large breeds" into ChatGPT, the answer does not come from your Instagram grid or your hero image. It comes from structured product data, review signals, and machine-readable catalog information. For the last decade, product discovery has been an auction-system bidding war. You pay Meta, you pay Google, you pay TikTok -- and the brand with the deepest pockets gets the most visibility. Creative quality and targeting help, but the underlying mechanic is spend-to-reach. AI commerce inverts that. There is no ad auction (for now). There is no bid. ChatGPT does not show your product because you paid for placement (again, for now). It shows your product because its data -- structured attributes, review signals, third-party citations, schema markup -- matched what the shopper asked for better than the next brand's data did. Data completeness is a different kind of competitive advantage. Not budget, not brand; *data completeness and merit*. And the uncomfortable part: "merit" here does not mean you have to have the best product. It means having the *best-structured data* about your product. A competitor with an objectively worse product but cleaner metafields, stronger review volume, and better schema markup will outrank you in AI recommendations. And that is already happening. The brands with the cleanest data are the ones who will dominate space on this new "retail shelf". This playbook covers the exact steps to audit, fix, and optimize your product catalog so AI agents recommend your brand first. *** ## How to Enable Agentic Storefronts If you are already on Shopify, Agentic Storefronts is likely active on your store right now. Shopify enabled it by default for all eligible US-based stores. Check your Shopify Admin under Settings, Sales Channels, Agentic Storefronts to confirm. **If you are on BigCommerce, Adobe Commerce/Magento, SAP Commerce Cloud, Agentforce Commerce, or custom builds, this still applies to you.** Shopify released the Agentic Plan on March 24. It is free, and you only pay when a product is sold (at standard Shopify Payments processing fees). You do not need to migrate your store or change platforms. Setup takes about 20 minutes: 1. Sign up for the [Agentic Plan](https://www.shopify.com/agentic-plan) via Shopify 2. Upload your product catalog to Shopify Catalog (it auto-enriches and categorizes your data) 3. Publish your store policies 4. Verify your existing domain via TXT DNS record 5. Provide external product URLs for each product For ChatGPT specifically, only product discovery setup is needed -- buyers complete purchases on your own store. For Google AI Mode and Copilot, full checkout configuration is required including shipping zones, tax collection, and payment setup through Shopify. One critical detail: Agentic Storefronts on the Agentic Plan are deactivated by default. You must manually activate them. The opposite of regular Shopify stores. Read that again: Shopify is positioning itself as the infrastructure layer for all AI commerce, not just its own merchants. A BigCommerce brand can surface products in ChatGPT through Shopify Catalog without touching their existing store. Whatever platform you are on, this channel is open. Currently US-based stores selling to US customers only. *** ## How AI Product Recommendations Actually Work To know how the product-recommendation engine works, you need to understand the infrastructure. Three layers determine whether AI agents surface your product or your competitor's. **Layer 1: Shopify Catalog: the data backbone.** Shopify runs specialized LLMs to categorize products, extract attributes, consolidate variants, and cluster identical items. Product data, such as titles, descriptions, options, images, pricing, and inventory, refreshes in real-time across all connected AI platforms. Not a product feed. Shopify's enterprise team said it directly: "Most product data was built for humans browsing websites. There is a gap between what renders well for a human shopper and what an AI agent can actually digest." Data trapped in Liquid templates, JavaScript rendering logic, or custom display rules are invisible to AI agents. If your product specs live only in rich HTML descriptions, they do not exist in this system. **Layer 2: Universal Commerce Protocol (UCP).** An open standard co-developed by Shopify and Google, endorsed by 20+ retailers and payment networks including Walmart, Target, Etsy, Visa, Mastercard, American Express, and Stripe. UCP establishes a common language for AI agents, merchants, and payment processors. It supports REST, GraphQL, JSON-RPC, Agent2Agent (A2A), and Model Context Protocol (MCP) transport protocols. The spec is open-source at [ucp.dev.](http://ucp.dev) Separately, OpenAI developed its own Agentic Commerce Protocol (ACP) with Stripe, focused on ChatGPT commerce specifically. Complementary but distinct, and worth tracking if you are building a platform-specific strategy. **Layer 3: The Knowledge Base App.** The lever most brands do not know about. You upload FAQs, policies, and brand voice guidelines. AI agents access this content during shopping conversations. Consumers never see it, but it directly shapes how AI represents your brand. \$50/month on standard plans. Free for Shopify Plus. This is the primary way you control how AI talks about your products. The full breakdown on how to populate it is covered later in this playbook. *** ## The Two-Tier Checkout System (And What It Breaks) Most of the coverage on Agentic Storefronts glosses over this, which is a disservice to merchants. There is not one checkout experience. There are two completely different ones, depending on which AI platform the buyer uses. Your optimization approach, margin math, and tracking setup need to account for both. **ChatGPT routes buyers to your own checkout.** Mobile: in-app browser inside ChatGPT. Desktop: links out to your store in a new tab. Best outcome for brands -- all your checkout customizations carry over. Upsells, cross-sells, checkout blocks, payment methods, loyalty widgets, branded flows. All of it. This happened after OpenAI pulled back its "Instant Checkout" in March 2026. Per CNBC, only about 30 merchants were ever on it, and users browsed but rarely bought in-chat. OpenAI said the feature "did not offer the level of flexibility that we aspire to provide." **Google AI Mode, Gemini, and Microsoft Copilot use a Shopify-powered built-in checkout** inside the AI platform itself. The buyer never leaves the conversation. Smoother for simple products -- but it strips out a lot of merchant functionality: * Product subscriptions do not work * Product bundles do not work * Checkout blocks load only those "essential to completing checkout" -- upsells, cross-sells, custom fields, quantity changes, and product addition blocks may not display * Local delivery, pickup in store, and pickup points are not supported * Custom pixels and GA4 do not fire -- only server-side events * Accelerated checkout supports only a subset of activated options * Automatic discounts and discount codes do work For brands with checkout-driven AOV strategies -- upsells, bundles, post-purchase offers -- the built-in checkout effectively amputates the bottom of the funnel. So if your growth strategy depends on a carefully tuned checkout or post-purchase experience, this will probably ring the alarms for you. But wait, there's another casualty: **Klaviyo behavioral tracking is dead for agentic shoppers across both checkout models.** Browse abandonment, cart abandonment, checkout abandonment -- none of those flows can trigger for AI-sourced sessions. Full breakdown in the Klaviyo section below. **The fee structure is split across platforms.** ChatGPT reportedly charges a 4% fee on completed sales, additive to standard \~2.9% Shopify Payments processing. That means the total cost for a ChatGPT-sourced order could approach \~7% when combining subscription, transaction, and processing fees. Google AI Mode and Microsoft Copilot currently charge no additional fees. Your finance team needs to model this now -- before AI-sourced orders scale from single-digit percentages to something that actually shows up on the P\&L. > **Note:** You can run the numbers yourself, but the math changes per SKU and per platform -- and most brands do not realize that until margin has already walked out the door. We build channel-specific margin models that map fee structures against your actual catalog: AOV delta between AI-referred and direct shoppers, upsell/bundle revenue lost in built-in checkouts, and the break-even point where ChatGPT's 4% gets offset by higher conversion rates. If you want that modeled for your catalog before the volume arrives, reach out at **[amlan@madebydas.com](mailto:amlan@madebydas.com)**. *** ## The Attribution Black Hole This is the section you should forward to your analytics team. GA4 and every client-side pixel are blind to agentic storefront checkouts that happen inside the built-in checkout (Google, Copilot). The purchase happens inside the AI platform's interface. Not on your site. No browser environment. No page where JavaScript tracking scripts fire. **What does NOT work for built-in checkouts:** | Tracking Tool | Status | | ---------------------------------------------------- | -------------------------------------------------------------- | | GA4 (client-side gtag.js) | No revenue data appears in GA4 reports | | Meta Pixel | Does not fire. Agentic purchases invisible to Meta Ads Manager | | TikTok Pixel, Pinterest Tag, Snapchat Pixel | Do not fire | | Triple Whale pixel | Does not fire. Orders appear as "unattributed" or "direct" | | Pre-checkout behavioral events | No page\_view, product\_viewed, add\_to\_cart, begin\_checkout | | UTM parameters and click IDs (fbclid, gclid, ttclid) | Not captured | | Cookie-based retargeting audiences | Cannot be built from agentic shoppers | **What DOES work:** | Tracking Tool | Status | | -------------------------------------------------- | -------------------------------------------------------------------------------------------------------------- | | Shopify Admin channel attribution | Every agentic order appears with originating AI channel labeled (ChatGPT, Google AI Mode, Copilot, Perplexity) | | Shopify Analytics reports | "Total sales by referrer" shows ChatGPT attribution; channel-specific filtering available | | Server-side pixel events | "Purchase started" and "Purchase completed" events fire | | Shopify webhooks | Standard order creation/update webhooks fire for all orders regardless of channel | | Meta Conversions API (CAPI) | If configured, purchase events sent server-side -- but without campaign-level attribution | | Google Enhanced Conversions / Measurement Protocol | Purchase events can reach GA4 via server-side API if configured | Even with server-side tracking configured, attribution quality is degraded. No click IDs to tie a purchase back to a specific ad. Server-side events tell platforms a conversion happened but cannot connect it to a campaign, so AI-sourced orders will inflate "organic" or "direct" in your multi-touch models -- which distorts ROAS calculations across every channel you run. ChatGPT is the exception. Because buyers complete checkout on your actual store (via in-app browser), client-side tracking is more likely to fire. But the in-app browser environment can still cause inconsistencies with cookie handling and third-party script loading. Test it, test it again, get your analytics team to validate it. Do not make assumptions here. **What this means for your ad spend.** If you run paid media (Meta, Google, TikTok, anything) AI-sourced orders are already messing with your attribution whether you have touched agentic commerce or not. Every AI-driven purchase that lands as "direct" or "organic" makes those channels look better than they are and makes your paid campaigns look worse. The brands that build an internal measurement framework for AI channels now will have a real edge once the volume scales. **The tool gap.** Triple Whale, Northbeam, Elevar, Littledata -- none have announced agentic commerce features. None have published guidance on how to handle AI-sourced orders in their models. The only purpose-built solution out there right now is CAPI Shield (from Stack Architect), a free tool that implements Meta CAPI and Google Enhanced Conversions server-side, aimed specifically at this gap. Beyond that, Shopify Admin is it. That is your only reliable source for AI channel performance data. Nobody has published a framework for reconciling Shopify Admin AI attribution data with GA4 or Meta. No guidance on calculating true CAC/ROAS for AI channels. No framework for handling AI-attributed orders in multi-touch models. Wide open. The brands that figure this out internally first get a compounding advantage that is hard to catch. > **Note:** No third-party tool solves this end-to-end, so we built the framework ourselves. For our clients that looks like: server-side tracking (Meta CAPI + Google Enhanced Conversions) configured before AI orders start flowing, a reconciliation layer that cross-references Shopify Admin AI channel data against GA4 and Triple Whale to isolate the attribution drift, and a weekly reporting cadence that pulls AI channel performance out of the "organic/direct" bucket it currently hides in. If your attribution stack is not ready for this, it is worth a conversation -- **[amlan@madebydas.com](mailto:amlan@madebydas.com)**. *** ## The Three AI Commerce Models, And Why Your Strategy Must Differ Per Platform These platforms do not all work the same way. Three models have emerged, and your optimization approach should differ for each. **Model 1: Discovery-first (ChatGPT, Perplexity).** ChatGPT now runs "Shopping Research" -- deep product comparisons powered by its Agentic Commerce Protocol, with ChatGPT memory for personalization. Purchases redirect to your site. Perplexity does "Buy with Pro" and "Instant Buy" with one-click checkout through Shopify. Both are research assistants, not transaction platforms. ChatGPT has 800 million weekly active users. Perplexity answers 125 million+ questions per week. Priority: product data completeness and off-site authority. Structured data and third-party citations are what determine who gets recommended. **Model 2: Full-stack agentic checkout (Google).** The most ambitious version. Google AI Mode has a "Buy for me" button -- Google's agent confirms the purchase details, adds items to the merchant's cart, and completes checkout via Google Pay. The buyer never visits your site. Backed by the Shopping Graph's 50+ billion product listings with 2+ billion refreshed hourly. Virtual try-on, AI price tracking, even the ability for AI to call local stores. No fees to merchants. Priority: Google Merchant Center feed accuracy and server-side tracking. Most transformative model. Also the most threatening to brand control. **Model 3: Embedded commerce (Microsoft Copilot).** Copilot Checkout launched January 2026 with PayPal as primary payment. In-line purchase panels inside the conversation -- no redirect. Alexander Del Rossa reported 3x higher conversion rates in Brand Agent-assisted sessions. Priority: product data parity with your Shopify catalog. Copilot's market share is single-digit percentage versus ChatGPT's \~68%. But the conversion rates point to high intent from the users who are there. **Amazon is the outlier.** They have blocked third-party AI agents (including ChatGPT) from scraping their site, sued Perplexity, and are investing in Rufus (250+ million customers have used it) and their own "Buy for Me" agent for external merchants. If you sell on Amazon, your Amazon optimization and your AI commerce strategy are two parallel tracks. Where this gets practical: a brand with strong checkout-driven AOV should weight ChatGPT (where your checkout is preserved) over Google AI Mode (where it is not). A brand competing on price transparency might find Google's model works in their favor. *** ## Fix Your Product Data Foundation Highest-impact section in this playbook. AI agents cannot recommend what they cannot parse. And most product catalogs right now are optimized for human browsing, not machine consumption. ### Move Specifications Into Metafields The most common problem we see: product details buried in long-form product descriptions (HTML). Specs, certifications, materials, size/weight -- all formatted for visual display, completely invisible to AI parsing. Shopify metafields use standard namespaces (e.g., `custom.specs`, `custom.materials`) that AI agents read directly. Every specification that matters for a purchase decision needs to live in a metafield. Attributes vary by category. The principle does not: * **For any product:** dimensions, weight, materials, country of origin, warranty, care instructions * **For wellness/supplements/consumables:** active ingredients and concentrations, certifications (organic, cruelty-free, vegan, non-GMO), allergen information, usage instructions, shelf life, nutrition facts panel * **For fashion/apparel:** fabric composition, fit type, size chart data, care instructions, sustainability certifications * **For electronics:** technical specifications, compatibility requirements, battery life, connectivity * **For home goods:** dimensions, material composition, origin, assembly requirements, weight capacity * **For pet products:** ingredient sourcing, life stage suitability, breed size recommendations If the information matters for a purchase decision, it needs to be in a metafield -- **NOT** buried in a description block. ### Rewrite Product Titles for Conversational Queries 70% of shopping queries in ChatGPT are phrased as natural language questions -- not keyword searches. Your product titles need to match how people actually talk. A title like "Premium NC-500 TWS" tells an AI agent nothing. A title like "Noise-Canceling Earbuds for Travel -- 40hr Battery, Bluetooth 5.3" matches exactly how someone would ask ChatGPT for a recommendation. Same principle across every category. Kill the internal SKU names. Lead with benefits in natural language: * Before: "RevitaGlow Pro Serum 30mL" -- After: "Vitamin C Brightening Serum -- 30mL with 15% L-Ascorbic Acid" * Before: "CloudLife Elite King Set" -- After: "Cooling Memory Foam Mattress Topper -- King, 3-Inch Gel-Infused" * Before: "PawNourish Premium Formula" -- After: "Grain-Free Senior Dog Food -- Joint Support with Glucosamine, 30lb Bag" The second version in each case answers what the AI agent is actually trying to figure out: what does this product do, what is in it, and who is it for? ### Implement JSON-LD Schema Markup Required structured data for AI commerce visibility: 1. **Product** -- name, description, brand, SKU, image 2. **Offer** -- price, currency, availability, condition 3. **AggregateRating** -- average rating and review count 4. **Review** -- individual review content 5. **FAQ** -- common product questions and answers 6. **Breadcrumb** -- category hierarchy If your Shopify theme does not already output JSON-LD for these types, add them. This is table stakes for AI discoverability. ### Use the Shopify Catalog Mapping Tool Settings, Agentic Storefronts, Catalog Mapping in your Shopify Admin. If you have custom data structures, this is where the mapping happens -- it takes your existing product data and maps it to the standardized format AI platforms consume. Go through every mapping. Check that certifications, specs, and category taxonomy are correctly mapped. This is where silent errors hide. ### Google's New GenAI Attributes Google added dozens of new data attributes built for conversational commerce -- product comparisons, use-case matching, contextual recommendations. These go well beyond traditional product feed attributes. Your Google Merchant Center feeds need to be complete, accurate, and refreshed daily. Google AI Overviews pull directly from GMC data. Stale feed = invisible in Google's AI surfaces. > **Note:** We have done this migration for brands with 500+ SKUs -- HTML description blocks into structured metafields, JSON-LD schema markup, Catalog Mapping verification, Google Merchant Center feed reconciliation. One client went from invisible in ChatGPT to top 3 for their primary category searches within 6 weeks. The only change was the metafield migration; just good data plumbing. And the single highest-ROI action in this entire playbook. If your catalog needs this, reach out at **[amlan@madebydas.com](mailto:amlan@madebydas.com)**. *** ## The Knowledge Base App, Your Brand Control Lever Most brands do not know this exists. That is the opportunity. The Knowledge Base App lets you upload content that AI agents pull from during shopping conversations. Consumers never see it. But it shapes every recommendation, comparison, and product description AI generates about your brand. ### What to Upload 1. **Brand positioning statement** -- Who you are, what you stand for, how you differentiate from competitors. Write this in plain language, not marketing copy. AI agents parse information, not vibes. 2. **Product FAQs** -- The 10-15 questions your customer support team answers most frequently. Format as clean question-answer pairs. These directly feed AI responses when shoppers ask about your products. 3. **Ingredient/material sourcing** -- Where your inputs come from, how they are processed, what certifications apply. AI agents surface this in response to questions like "Is \[brand] actually organic?" or "Where does \[brand] source their materials?" 4. **Return and shipping policies** -- Clear, structured policies reduce purchase friction in AI conversations. 5. **Brand voice guidelines** -- How your brand should be described. Subtle but it works. If you position as premium and clinical, say so. If you are approachable and community-driven, say so. The AI adjusts its tone when discussing your products accordingly. ### What Not to Upload Do not upload marketing fluff, aspirational mission statements, or buzzword-heavy positioning decks. AI agents parse for factual density. Vague language gets ignored. Do not upload confidential pricing strategies, margin data, or competitive intelligence. This content is accessible to the AI systems -- treat it as semi-public. ### The Test After you populate the Knowledge Base, give it 48 hours to propagate. Then re-run the 15-minute audit from the Quick Start section. Compare before and after. The shift is usually obvious -- especially on questions about positioning and how AI differentiates you from competitors. *** ## Off-Site Authority Signals Product data gets you into the conversation. Off-site authority is what decides whether AI recommends you or the other brand. Princeton published a study (ACM SIGKDD 2024) showing that targeted Generative Engine Optimization strategies increased visibility in AI-generated responses by up to 40%. AI platforms weight third-party citations heavily. Here is what moves the needle: ### Review Site Presence AI agents pull from review aggregators, editorial reviews, and community feedback. Perfect schema markup with thin review coverage and negative Reddit sentiment loses to a competitor with stronger off-site signals. Every time. Prioritize reviews on your own site (first-party), Amazon (if applicable), and category-specific platforms. Volume matters more than you think. 2,000 reviews outranks 50 reviews in AI recommendation logic regardless of average rating. ### Reddit and Community Signals Reddit has become a primary citation source for AI models. Search Reddit for your brand name and product categories. If the conversation is thin or negative, that is directly affecting your AI visibility. You cannot manufacture Reddit sentiment. But you can ensure your products are being discussed in relevant subreddits by engaging authentically in community conversations and ensuring your PR and influencer strategy includes Reddit-native creators. ### Editorial Coverage Category publications, ingredient deep-dives from editors, curated "best of" lists -- these all feed the recommendation engines. One well-placed editorial mention can shift how AI frames your brand in competitive comparisons. The move: check which editorial sources AI agents are currently citing for your category. Then prioritize earned media in those specific publications. Work backward from where the AI is already looking. *** ## Klaviyo and Email/SMS: What Works and What Doesn't If you run Klaviyo, read this carefully. Klaviyo captures agentic orders -- but only partially. The Shopify integration syncs via API and webhooks, not client-side pixels. So "Placed Order" and "Ordered Product" events fire for agentic purchases. Post-purchase flows work: order confirmation, shipping updates, review requests, cross-sell sequences, win-back campaigns. All normal. **Pre-purchase behavioral events are gone.** Active on Site, Viewed Product, Added to Cart, Checkout Started -- none fire for agentic shoppers. Browse abandonment, cart abandonment, checkout abandonment flows cannot trigger for AI-sourced sessions. This means that your post-purchase flows are the only re-engagement levers for AI-sourced customers. If a big chunk of your Klaviyo revenue comes from abandonment flows, the AI channel will look like it underperforms on a per-session basis -- even though AI-referred shoppers convert at higher rates on first visit. Your post-purchase sequences need to be strong. Review request timing, cross-sell logic, win-back cadence -- they carry more weight when you cannot warm up the customer before they buy. Klaviyo has not announced anything to address this. *** ## Technical Readiness Checklist Seven items. Most take under an hour. All affect whether AI agents can find, index, and correctly attribute your product data. 1. **robots.txt** -- Confirm that OAI-SearchBot (OpenAI's crawler) is not blocked. Many Shopify stores inherited restrictive robots.txt rules from SEO agencies that block AI crawlers by default. 2. **llms.txt** -- An emerging specification for AI crawlers. Place a structured file at your site root that tells AI agents what content is available and how to access it. Early adopters are seeing measurable indexing advantages. 3. **IndexNow protocol** -- Real-time content change notifications. When you update a product price, add a new SKU, or change availability, IndexNow pushes that update to search engines and AI platforms immediately instead of waiting for the next crawl cycle. 4. **Google Merchant Center** -- Ensure feeds are complete, accurate, and refreshed daily. Google AI Overviews pull directly from GMC data. 5. **Server-side tracking** -- Configure Meta Conversions API (CAPI) and Google Enhanced Conversions before AI-sourced orders start flowing. Client-side pixels (GA4, Meta Pixel, Triple Whale) are completely blind to agentic checkout orders. This is not optional -- it is the only way to measure AI channel performance outside of Shopify Admin. 6. **Klaviyo webhook verification** -- Confirm that Klaviyo's Shopify integration is syncing via API and webhooks. Post-purchase flows (order confirmation, shipping, review requests) will work for AI-sourced customers. Pre-purchase flows (browse abandonment, cart abandonment) will not -- plan accordingly. 7. **Shopify Admin attribution check** -- Navigate to Analytics, Reports. Filter by AI channel. Confirm that the referrer attribution is correctly labeling ChatGPT, Google AI Mode, and Copilot orders. This is your single source of truth for AI channel performance. *** ## The Weekly AI Channel Review None of this is "set and forget"; this is a channel; manage it like one. ### What to Track Weekly 1. **AI channel order count** -- Shopify Admin is the single source of truth. Filter "Total sales by referrer" to see ChatGPT, Google AI Mode, and Copilot attribution. 2. **AI channel AOV** -- AI-referred shoppers show higher average order values than direct site traffic. Track this separately. If AOV from AI channels runs 20-30% higher, it changes the margin math on those ChatGPT fees. 3. **Attribution drift** -- Monitor your "direct" and "organic" attribution in GA4 and your multi-touch attribution tool. If those numbers are climbing without a clear cause, AI-sourced orders may be inflating them. Cross-reference against Shopify Admin AI channel data. 4. **Competitive positioning shifts** -- Re-run 2-3 queries from your audit monthly. Monitor whether your brand is gaining or losing position relative to competitors. 5. **Knowledge Base content freshness** -- Update FAQs and product information as your catalog changes. Stale data degrades AI representation over time. ### The Cadence **Week 1-4 after launch:** Daily monitoring. Check order attribution and scan for any data mapping issues. **Month 2 onward:** Weekly 15-minute review. Pull AI channel metrics, update Knowledge Base if products changed, run one competitive query. **Quarterly:** Full audit re-run. All five queries across ChatGPT, Gemini, and Perplexity. Compare against your baseline. Document improvement or regression. Build this cadence now while AI-sourced orders are still small. The operational muscle you develop at low volume is what lets you scale the channel when the volume arrives. *** ## What Comes Next Everything in this playbook is implementable. Metafield migration, Knowledge Base population, the technical checklist -- a focused team can get the core optimizations done in a week. Maintaining it is where brands stall. Product data needs updating as catalogs change. The Knowledge Base needs refreshing as new products launch. Off-site authority compounds slowly -- months, not weeks. Attribution reconciliation between Shopify Admin, GA4, and your multi-touch tools needs ongoing attention as both the AI platforms and the tracking tools evolve. And the AI platforms are updating their recommendation algorithms weekly. What works in Q2 may need adjustment in Q3. *** ## Two Ways to Move Forward **Option 1: Run this playbook yourself.** Everything in this guide works. Start with the 15-minute audit, move to metafield migration, populate the Knowledge Base, configure server-side tracking. You will see measurable improvement in AI recommendations within 2-4 weeks. Where most brands hit a wall: maintaining the system across a growing catalog while simultaneously managing paid media, creative production, attribution reconciliation, and the 15 other things competing for your growth team's attention. > **Note:** **Option 2: Let us handle it.** > > This is what DAS does -- metafield migration across your full catalog, Knowledge Base population and ongoing management, server-side tracking configuration, AI channel attribution framework, weekly performance monitoring, and the competitive audit cadence that keeps you ahead of positioning shifts. The full stack, managed as a channel -- the same way you manage paid media, except nobody on your team has to learn a new discipline. > > If your brand is doing \$35M+ and you want AI commerce handled as a managed capability, not a side project, we should talk. [amlan@madebydas.com](mailto:amlan@madebydas.com) # The Claude Agents Playbook: 18 AI Agents for Ecommerce Operators Source: https://intelligence.madebydas.com/playbooks/claude-agents-playbook 18 pre-configured Claude agent setups across 6 operating stacks for ecommerce operators — SKU margin ranking, ROAS reconciliation, at-risk customer detection, cohort LTV builds, contribution margin by channel, and more. Each agent includes an Agent Brief, Input Spec, and Output Spec. Runs in Claude Projects with zero code. Built by DAS — The Audience Development Agency. # The Claude Agents Playbook: 18 AI Agents for Ecommerce Operators ## What This Is Claude agents for ecommerce are pre-configured AI specialists that run recurring analysis workflows — SKU margin ranking, ROAS reconciliation, at-risk customer detection, cohort LTV builds, and contribution margin by channel — using data exported from Shopify and Klaviyo. Each agent has three components: an Agent Brief (system prompt configured once in Claude Projects), an Input Spec (the exact CSV export to provide), and an Output Spec (the formatted analysis returned). This playbook covers 18 agents across 6 operating stacks that run with zero code in Claude Projects. ## How It Works Every agent in this playbook follows the same three-component structure: **Agent Brief** — A system prompt you configure once in a Claude Project. It tells Claude its role, the analytical framework it should apply, the specific outputs you expect, and the format of the response. Once set, it persists across every session in that project. **Input Spec** — The exact data export you provide at the start of each session. Usually a CSV or spreadsheet from Shopify or Klaviyo. The spec tells you exactly which fields to include and how to format them. **Output Spec** — The formatted analysis Claude returns. Designed to be actionable: a table with the ranked finding, the implication, and the recommended action — not a prose summary. This structure is repeatable. Once you build the agent brief once, you run the same analysis weekly by uploading the latest export and asking Claude to run the analysis. *** ## Stack 1: Customer Intelligence *The foundation. Run these before any other stack.* ### Agent 1: RFM Distributor **What it does:** Ingests a Shopify order export and segments every customer into RFM tiers. Outputs the size of each segment, the revenue contribution of each segment, and the percentage of total contribution margin each segment represents. **Agent Brief:** ``` You are an RFM segmentation analyst for a DTC ecommerce brand. When I upload a Shopify order export, score every customer on Recency (days since last order), Frequency (total order count), and Monetary Value (total spend). Assign each customer to one of six tiers: Champion, Loyal, Recent, Needs Attention, At Risk, Inactive. Output a summary table showing: segment name, customer count, % of total customers, total revenue, % of total revenue, estimated contribution margin %. Sort by revenue contribution descending. Flag if Champions represent less than 40% of revenue — this indicates a concentration problem. ``` **Input Spec:** Shopify orders export (CSV). Required fields: Customer ID, Order Date, Order Total, Discount Amount. **Output Spec:** RFM distribution table + concentration flag + top 3 action recommendations. *** ### Agent 2: Cohort LTV Builder **What it does:** Takes a Shopify order export grouped by acquisition month and builds 30, 60, 90, 180, and 365-day LTV curves for each cohort. Identifies which cohorts are compounding versus which are churning early. **Agent Brief:** ``` You are a cohort LTV analyst. When I upload a Shopify order export with acquisition dates, group customers by their first-purchase month (acquisition cohort). For each cohort, calculate cumulative revenue at 30, 60, 90, 180, and 365 days after first purchase. Output a cohort table with these milestones. Identify the top 3 cohorts by 365-day LTV and the bottom 3. For the top cohorts, flag what acquisition period they came from — this tells us which seasons or campaigns produced the best customers. ``` **Input Spec:** Shopify orders export (CSV). Required fields: Customer ID, First Order Date, All Subsequent Order Dates, Order Totals. **Output Spec:** Cohort LTV table by month + top/bottom cohort analysis + acquisition period flags. *** ### Agent 3: Champion Segment Profiler **What it does:** Takes the Champion segment from the RFM analysis and builds a behavioral profile: first product purchased, average time to second purchase, average order frequency, acquisition source patterns. **Agent Brief:** ``` You are a customer intelligence analyst. When I upload data for the Champion customer segment, identify behavioral patterns that distinguish this segment: most common first product purchased, average days between first and second purchase, average order frequency, geographic concentration, and any Klaviyo flow or campaign tags present. Output a Champion Profile Card with these findings. This output becomes the brief for acquisition creative and paid media lookalike audiences. ``` **Input Spec:** Shopify orders export filtered to Champion segment (from Agent 1 output). Fields: Customer ID, First Product, Order Dates, Geography, Klaviyo source tags if available. **Output Spec:** Champion Profile Card formatted as a one-page brief. *** ## Stack 2: Margin Intelligence *Where the P\&L conversation starts.* ### Agent 4: SKU Margin Ranker **What it does:** Ingests a product margin spreadsheet and ranks SKUs by contribution margin — not gross margin. Flags SKUs where high revenue is masking negative contribution margin after accounting for returns, discounts, and variable costs. **Agent Brief:** ``` You are a SKU-level margin analyst. When I upload a product data export with revenue, COGS, return rate, average discount applied, and shipping cost per unit, calculate contribution margin per SKU: Revenue × (1 - return rate) - COGS - average discount - shipping cost. Rank all SKUs by contribution margin descending. Flag any SKU where gross margin is positive but contribution margin is negative. These are margin traps. Output a ranked table with a "Margin Trap" flag column and recommended actions for the bottom 10% of SKUs. ``` **Input Spec:** Product/SKU spreadsheet. Required fields: SKU, Revenue, Units Sold, COGS, Return Rate %, Average Discount %, Shipping Cost. **Output Spec:** SKU margin ranking table + Margin Trap flags + action recommendations. *** ### Agent 5: Channel Margin Comparator **What it does:** Compares contribution margin per customer by acquisition channel — paid social, paid search, email, organic, creator — to identify which channels produce margin-positive cohorts versus volume-only cohorts. **Agent Brief:** ``` You are a channel attribution analyst focused on contribution margin, not ROAS. When I upload a report showing customers by acquisition channel with their subsequent purchase history and estimated margin, calculate contribution margin per acquired customer by channel at 90 and 365 days. Rank channels by 365-day contribution margin per customer. Flag any channel where volume is high but 365-day contribution margin per customer is below the brand average. This indicates a volume-without-margin problem. ``` **Input Spec:** Customer acquisition channel export + order history. Fields: Customer ID, Acquisition Channel, Acquisition Date, All Order Dates, Order Totals, Discount %. **Output Spec:** Channel margin comparison table + 90/365-day views + volume-without-margin flags. *** ### Agent 6: Discount Leakage Identifier **What it does:** Identifies where discount codes are being applied to customers who would likely have purchased without them — specifically, Champions and Loyal customers receiving promotional discounts. **Agent Brief:** ``` You are a promotional efficiency analyst. When I upload an order export with RFM segment and discount code applied, identify orders where Champion or Loyal segment customers received a discount of 15% or more. Calculate the total margin lost from unnecessary discounting of these segments. Output a Discount Leakage Report: total orders, total margin lost, and the specific discount codes driving the most leakage. These are the campaigns to restructure or eliminate. ``` **Input Spec:** Order export with RFM segment labels and discount codes applied. Fields: Customer ID, RFM Segment, Order Total, Discount Code, Discount Amount. **Output Spec:** Discount Leakage Report with total margin lost + code-level breakdown + restructuring recommendations. *** ## Stack 3: Retention Intelligence *The 90-day window where most brands lose.* ### Agent 7: At-Risk Customer Detector **What it does:** Ingests Klaviyo customer data with purchase history and churn risk scores to identify which high-value customers are approaching the point where intervention becomes impossible. **Agent Brief:** ``` You are a retention intelligence analyst. When I upload a Klaviyo customer export with churn risk scores and LTV data, identify all customers who are: (1) Medium or High churn risk AND (2) have LTV above the brand average. Sort by LTV descending. For each customer, calculate the revenue at risk (estimated 12-month future value if retained vs. lost). Output an At-Risk Recovery Priority List with revenue at risk per customer and recommended intervention type: personal outreach for top 20, automated flow for next 80, standard winback for remainder. ``` **Input Spec:** Klaviyo customer export with churn risk scores and CLV. Fields: Customer ID, Email, CLV, Churn Risk (Low/Medium/High), Last Order Date, Total Orders. **Output Spec:** Priority list sorted by revenue at risk + intervention type recommendation per customer tier. *** ### Agent 8: Second-Purchase Optimizer **What it does:** Analyzes the first-to-second-purchase journey to identify the optimal intervention window and the products that most reliably trigger a second purchase. **Agent Brief:** ``` You are a post-purchase retention analyst. When I upload a Shopify order export for first-time buyers, calculate: (1) median days from first to second purchase for customers who did purchase again, (2) the first product purchased by customers who converted to second purchase at the highest rate, and (3) the first product purchased by customers who never purchased again. Output a Second Purchase Conversion Map: products ranked by repeat rate, optimal intervention window (days), and recommended post-purchase sequence timing. ``` **Input Spec:** Shopify orders export filtered to customers with 1–2 orders. Fields: Customer ID, First Order Date, First Product Purchased, Second Order Date (if exists), Second Product Purchased (if exists). **Output Spec:** Second Purchase Conversion Map with product-level repeat rates + timing recommendations. *** ### Agent 9: Winback Segment Builder **What it does:** Segments lapsed customers by value and recency to build differentiated winback strategies — personal outreach for high-value, automated for mid-value, suppression for low-value. **Agent Brief:** ``` You are a winback strategy analyst. When I upload a customer export of lapsed customers (no purchase in 90+ days), segment them using a 2×2 matrix: High Value / Low Value (relative to brand average LTV) × High Recency Risk / Low Recency Risk (last purchase 90–180 days vs. 180+ days). For each quadrant, recommend the appropriate intervention: personal outreach with exclusive offer, automated flow with progressive discount, one-touch email then suppress, or immediate suppression. Output the matrix with customer counts per quadrant and the specific intervention brief for each. ``` **Input Spec:** Customer export of lapsed customers. Fields: Customer ID, LTV, Last Order Date, Total Orders, Average Order Value. **Output Spec:** 2×2 winback matrix with customer counts + intervention brief per quadrant. *** ## Stack 4: Media Intelligence *Connect the ad account to the customer file.* ### Agent 10: Lookalike Seed Builder **What it does:** Outputs a clean, formatted Champion segment list ready for upload as a custom audience seed in Meta — with the LTV column Meta uses for value-based lookalikes. **Agent Brief:** ``` You are a paid media data analyst. When I upload a customer export with LTV data, filter to the top 2,000 customers by LTV. Format the output as a Meta-ready custom audience file with the following columns: email (hashed), phone (hashed if available), first name, last name, LTV (for value-based lookalike). Remove any customers who have unsubscribed from email. Output the formatted list and a summary: average LTV of seed list, LTV range, and recommendation for 1% vs. 2% lookalike based on seed size. ``` **Input Spec:** Customer export with LTV. Fields: Customer ID, Email, Phone, First Name, Last Name, LTV, Email Subscription Status. **Output Spec:** Meta-formatted CSV ready for upload + seed summary. *** ### Agent 11: ROAS Reconciler **What it does:** Compares platform-reported ROAS against Shopify revenue to quantify the attribution gap and identify which channel's self-reported numbers are most overstated. **Agent Brief:** ``` You are an attribution analyst. When I upload platform ROAS reports alongside Shopify revenue data for the same period, calculate: (1) total revenue claimed by all platforms combined, (2) actual Shopify revenue for the period, (3) the over-attribution gap (claimed minus actual), and (4) which platform's numbers contribute most to the gap. Output an Attribution Audit Report with the gap in dollar terms and percentage, and recommend whether to run a geo holdout or media mix model based on the size of the discrepancy. ``` **Input Spec:** Platform ROAS reports (Meta, Google, TikTok) + Shopify revenue export for same date range. Fields: Platform, Spend, Claimed Revenue, Clicks. Shopify: Total Revenue, Orders, Period. **Output Spec:** Attribution Audit Report with gap analysis + incrementality testing recommendation. *** ### Agent 12: Channel Mix Optimizer **What it does:** Uses cohort LTV data by acquisition channel to recommend a budget reallocation that optimizes for 365-day contribution margin rather than platform ROAS. **Agent Brief:** ``` You are a media budget allocation analyst. When I upload channel performance data with 365-day cohort LTV per acquisition channel, rank channels by 365-day contribution margin per dollar spent — not ROAS, not CAC. Calculate how budget would need to shift to maximize 365-day contribution margin given current total spend. Output a Budget Reallocation Recommendation with current allocation, recommended allocation, and the projected margin impact of the shift. ``` **Input Spec:** Channel spend + cohort LTV by channel. Fields: Channel, Spend (period), Customers Acquired, 90-day LTV per customer, 365-day LTV per customer, Average contribution margin %. **Output Spec:** Reallocation recommendation table + projected margin impact. *** ## Stack 5: Creative Intelligence *Brief from data, not instinct.* ### Agent 13: Review Mining Agent **What it does:** Processes a bulk export of customer reviews to extract the specific phrases, emotional triggers, and pain points that should appear in creative briefs. **Agent Brief:** ``` You are a customer voice analyst. When I upload a set of customer reviews, identify: (1) the top 15 emotional triggers and specific phrases customers use to describe product transformation, (2) the top 10 objections customers mention before or during their decision to purchase, (3) the top 5 pain points the product resolved, and (4) any phrases that appear verbatim 3+ times. Output a Creative Brief Input Document formatted for direct use in a campaign brief: customer language for headlines, body copy, and objection handling. ``` **Input Spec:** Customer reviews export (text file or CSV). Fields: Review Text, Star Rating, Date. **Output Spec:** Creative Brief Input Document with verbatim customer phrases, objection list, and usage recommendations. *** ### Agent 14: Segment Creative Strategist **What it does:** Takes the Champion Profile (from Agent 3) and the Review Mining output (from Agent 13) and builds a creative brief specific to the Champion segment for a given campaign objective. **Agent Brief:** ``` You are a performance creative strategist. When I provide (1) a Champion Segment Profile and (2) a Review Mining output, and (3) a campaign objective, write a creative brief for the Champion segment. The brief must include: the one insight about Champion customers that this campaign will leverage, the specific customer language to use (verbatim phrases from reviews), the objection to address, the offer or value proposition, and the format recommendation for each placement (feed, story, email, SMS). Do not use generic DTC creative conventions. Write from the Champion's actual vocabulary. ``` **Input Spec:** Champion Profile Card (Agent 3 output) + Creative Brief Input Document (Agent 13 output) + campaign objective statement. **Output Spec:** Formatted creative brief with insight, customer language, objection, value prop, and format recommendations per placement. *** ### Agent 15: Email Subject Line Generator **What it does:** Generates subject line variants calibrated to specific RFM segments — not generic A/B testing variants, but segment-specific language built from behavioral understanding. **Agent Brief:** ``` You are an email subject line specialist for RFM-segmented campaigns. When I provide the RFM segment being targeted, the campaign objective, and 5–10 verbatim customer phrases from the review mining analysis, generate 10 subject line variants calibrated to that segment's behavioral profile. Champion variants should emphasize exclusivity, recognition, and access. At-Risk variants should emphasize value and social proof. New customer variants should emphasize trust and education. Label each variant with the segment it's optimized for and the psychological trigger it uses. ``` **Input Spec:** RFM segment name, campaign objective, customer phrase list (from Agent 13). **Output Spec:** 10 subject line variants with segment label and psychological trigger annotation. *** ## Stack 6: Seasonal & Competitive Intelligence ### Agent 16: Seasonal Cohort Planner **What it does:** Analyzes two years of purchase history to identify which customer segments respond to seasonal promotions, what they respond to, and what the optimal timing is. **Agent Brief:** ``` You are a seasonal strategy analyst. When I upload two years of Shopify order data with RFM segment labels, identify: (1) which segments showed incremental purchase activity during BFCM, Valentine's Day, Mother's Day, and Back-to-School relative to their base purchase rate, (2) what the average discount applied was for orders in each segment during each seasonal period, (3) whether Champion customers had higher or lower discount sensitivity during seasonal periods. Output a Seasonal Response Map per segment with timing and offer recommendations for the next 12 months. ``` **Input Spec:** 2-year Shopify orders export with RFM segment labels. Fields: Customer ID, RFM Segment, Order Date, Order Total, Discount Code, Discount Amount. **Output Spec:** Seasonal Response Map with segment-level response rates, timing recommendations, and offer structure by segment. *** ### Agent 17: Competitive Review Analyzer **What it does:** Processes competitor reviews to surface unmet customer needs, product gaps, and messaging angles the brand can own. **Agent Brief:** ``` You are a competitive intelligence analyst. When I upload customer reviews from [Competitor], identify: (1) the top 5 product attributes competitors are praised for (potential parity requirements), (2) the top 5 complaints that represent unmet needs (potential differentiation opportunities), (3) any mentions of alternative brands customers considered, and (4) the emotional language customers use when they are disappointed versus delighted. Output a Competitive Gap Map with the three highest-priority differentiation opportunities and recommended messaging angles. ``` **Input Spec:** Competitor reviews export. Fields: Review Text, Star Rating. **Output Spec:** Competitive Gap Map with ranked opportunities and messaging recommendations. *** ### Agent 18: Weekly Intelligence Briefer **What it does:** Takes a weekly data dump from Klaviyo and Shopify and produces a one-page intelligence brief with the 3 most important findings and the 3 recommended actions. **Agent Brief:** ``` You are a weekly marketing intelligence analyst. When I upload this week's Klaviyo segment performance report and Shopify order summary, compare to the prior week's data (which I will also provide) and identify: (1) any segment that changed size by more than 10% week-over-week (both growth and decline), (2) any change in Champion segment revenue share, (3) any SKU with an unusual spike or drop in units sold. Output a one-page Weekly Intelligence Brief: Top 3 Findings + Top 3 Actions. Keep it under 300 words. This goes in the rolling Customer Intelligence Brief. ``` **Input Spec:** Current week Klaviyo segment report + current week Shopify order summary + prior week equivalents for comparison. **Output Spec:** One-page Weekly Intelligence Brief with 3 findings + 3 actions, under 300 words. *** ## How to Deploy Create one Claude Project per stack (or one master project for all agents). In each project, paste the Agent Brief for each agent into the project instructions. Claude will remember this context across every session. Set up recurring exports from Shopify and Klaviyo that match each agent's Input Spec. Most can be set to run automatically on a weekly schedule. Upload the relevant export and type: "Run the \[Agent Name] analysis on this data." Claude will return the Output Spec formatted analysis. Create a Notion page or Google Doc. After each agent run, paste the key findings. Over time, this becomes the institutional intelligence that makes every subsequent decision smarter. Click **Open in Claude** above to start configuring any of these agents with your own data. Claude will have this full playbook as context and can help you customize the Agent Briefs for your specific Shopify and Klaviyo setup. # Claude Cowork Plugins: The Complete Implementation Guide Source: https://intelligence.madebydas.com/playbooks/claude-cowork-plugins The complete guide to Claude Cowork plugins — modular extensions that transform Claude from a conversational AI into a specialized automation platform. 10 real-world workflows for HR, finance, legal, sales, and operations. ## Before You Dive In This guide is intentionally dense. If you need a custom workflow built, a skill developed, or just a second set of eyes — reach out. Amlan Das, Founder — DAS Audience Development [amlan@madebydas.com](mailto:amlan@madebydas.com) *** ## What Are Claude Cowork Plugins? ### Overview Claude Cowork plugins are modular extensions that transform Claude from a conversational AI into a specialized automation platform tailored to specific roles, teams, and workflows. Launched on **January 30, 2026**, the plugin system enables users to bundle multiple customization types into single installable packages. Plugins bring the same agentic architecture from Claude Code (the developer-focused tool) to Claude Cowork (the knowledge work productivity tool), without requiring terminal access or coding expertise. ### Core Concept Instead of configuring Claude from scratch for each task, plugins provide ready-made bundles containing: * **Slash Commands**: Quick shortcuts for common workflows (e.g., `/rfp:extract`, `/legal:discover`) * **Sub-agents**: Specialized AI workers with isolated contexts for parallel task execution * **MCP Servers**: Connections to external tools and data sources (Slack, Salesforce, Google Workspace) * **Hooks**: Automation triggers at lifecycle events (e.g., auto-format after file edits) * **Skills**: Instructions Claude reads automatically based on task context **Real-World Analogy**: Think of plugins as "app stores" for Claude — instead of using a generic assistant, you install plugins that turn Claude into a specialized RFP analyst, financial auditor, or legal document reviewer. ### Key Capabilities | Capability | Description | Example | | -------------------------- | --------------------------------------------- | --------------------------------------------------------------- | | **Direct File Access** | Read/write local files without manual uploads | Process 200 resumes in a folder into a ranked Excel spreadsheet | | **Sub-agent Coordination** | Break complex work into parallel workstreams | 5-year financial analysis: one sub-agent per fiscal year | | **Professional Outputs** | Generate formatted deliverables | Excel with formulas, PowerPoint decks, formatted Word docs | | **Long-Running Tasks** | Extended execution without timeouts | Analyze 50 contracts for risk clauses (30+ minute task) | | **Browser Integration** | Control web browsers via Chrome extension | Scrape competitor pricing from websites | **Evidence**: NASA used custom Claude plugins to generate Mars rover driving instructions, achieving **50% time savings** on a previously manual process. *** ## System Requirements and Setup ### Requirements | Requirement | Specification | Notes | | -------------------- | -------------------- | ------------------------------------------- | | **Operating System** | macOS only | Windows support coming (timeline TBD) | | **Application** | Claude Desktop app | Not available on web or mobile | | **Subscription** | Paid plan required | Pro (\$20/month), Max, Team, or Enterprise | | **Internet** | Active connection | Required throughout session | | **Permissions** | Folder access grants | You control which folders Claude can access | ### Installation Process #### Step 1: Install Claude Desktop ```text theme={null} 1. Download Claude Desktop for macOS from https://claude.com/download 2. Install the application 3. Sign in with your Claude account (Pro, Max, Team, or Enterprise) ``` #### Step 2: Access Cowork ```text theme={null} 1. Open Claude Desktop 2. Look for the mode selector showing "Chat" and "Cowork" tabs 3. Click the "Cowork" tab to switch to Tasks mode ``` #### Step 3: Grant File Access (First Time) ```text theme={null} 1. Cowork will prompt you to grant folder access 2. Best practice: Create a dedicated work folder first mkdir ~/cowork-workspace 3. Click "Choose Folder" and select ~/cowork-workspace 4. Grant permissions when prompted ``` **Security Note**: Claude can only access files in folders you explicitly grant. Nothing is accessible by default. #### Step 4: Verify Setup Test with a simple task: ```text theme={null} Organize the files in this folder by type and create a summary report. ``` If Claude responds with a plan and begins working, you're set up correctly. *** ## Plugin Architecture Deep Dive ### The Five Plugin Components Plugins bundle up to five distinct extension types, each serving a specific automation purpose: ```text theme={null} my-plugin/ ├── .claude-plugin/ │ └── plugin.json # Required: Plugin manifest ├── commands/ # Slash commands (Markdown files) │ ├── extract-rfp.md │ └── analyze-contracts.md ├── agents/ # Sub-agent definitions │ ├── requirement-extractor.md │ ├── risk-analyzer.md │ └── financial-parser.md ├── skills/ # Auto-invoked instructions │ ├── legal-taxonomy/ │ │ └── SKILL.md │ └── financial-ratios/ │ └── SKILL.md ├── hooks/ │ └── hooks.json # Lifecycle automation ├── .mcp.json # External tool integrations └── scripts/ # Helper utilities └── format-output.sh ``` *** ### Component 1 — Slash Commands **Purpose**: Reusable text-based shortcuts for common workflows **File Format**: Markdown files in `commands/` directory **Invocation**: Manual (`/plugin-name:command-name`) #### Example: /rfp:extract ```markdown theme={null} --- description: Extract all requirements from RFP documents arguments: --- # RFP Requirements Extraction Analyze all RFP documents in the specified folder and extract: 1. **Functional Requirements**: Features, capabilities, specifications 2. **Technical Requirements**: Platforms, integrations, security standards 3. **Compliance Requirements**: Certifications, regulatory mandates 4. **Evaluation Criteria**: Scoring rubrics, decision factors, weighting For each requirement: - Assign unique ID (REQ-001, REQ-002...) - Categorize as Mandatory or Optional - Note RFP page number for citation Output: Excel workbook with requirements matrix, coverage tracking formulas, and risk flags. Arguments provided by user: $ARGUMENTS ``` **How It Works**: * User types `/rfp:extract /path/to/rfp-folder/` * `$ARGUMENTS` placeholder is replaced with `/path/to/rfp-folder/` * Claude executes the workflow described in the command *** ### Component 2 — Sub-agents **Purpose**: Specialized AI workers with isolated contexts for parallel execution **Architecture**: Each sub-agent has separate context window, custom system prompt, scoped tool permissions **File Format**: Markdown with YAML frontmatter in `agents/` directory #### Example: Requirement Extractor Sub-agent ```yaml theme={null} --- name: requirement-extractor description: Extract structured requirements from RFP sections model: sonnet-4.5 tools: [Read, Grep] permissions: file_access: read_only --- You are a requirements extraction specialist. Your role is to: 1. Read RFP sections assigned to you 2. Identify all requirements (functional, technical, compliance) 3. Extract exact requirement text with page citations 4. Classify each as mandatory vs. optional 5. Return structured JSON to orchestrator DO NOT: - Make assumptions about unstated requirements - Editorialize or paraphrase requirement language - Access files outside your assigned scope Return findings in this format: { "requirements": [ { "id": "REQ-001", "category": "Functional", "text": "System must support SSO via SAML 2.0", "mandatory": true, "page": 23 } ] } ``` **Key Innovation**: Sub-agents report only summary findings back to main orchestrator, preventing context pollution when processing large documents. *** ### Component 3 — MCP Servers (Model Context Protocol) **Purpose**: Connect Claude to external tools and data sources **File Format**: `.mcp.json` at plugin root #### Example: CRM Integration ```json theme={null} { "mcpServers": { "salesforce": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-salesforce"], "env": { "SALESFORCE_USERNAME": "${SALESFORCE_USERNAME}", "SALESFORCE_TOKEN": "${SALESFORCE_TOKEN}" } }, "google-sheets": { "command": "${CLAUDE_PLUGIN_ROOT}/servers/sheets-server.js", "args": ["--credentials", "${GOOGLE_CREDENTIALS_PATH}"], "cwd": "${CLAUDE_PLUGIN_ROOT}" } } } ``` #### Available Integrations (via MCP ecosystem) * **Productivity**: Slack, Asana, Linear, Jira, Notion * **Cloud Storage**: Google Drive, Dropbox, Box * **CRM**: Salesforce, HubSpot * **Development**: GitHub, GitLab, Sentry * **Data**: PostgreSQL, MongoDB, Redis * **Analytics**: Amplitude, Mixpanel, Google Analytics **How It Works**: * MCP servers expose tools (functions), resources (data), and prompts (templates) through standardized JSON-RPC interface * Claude invokes these tools seamlessly alongside built-in capabilities * Servers start automatically when plugin is enabled *** ### Component 4 — Hooks **Purpose**: Lifecycle event automation (validation, enrichment, blocking) **File Format**: `hooks/hooks.json` #### Available Events * `PreToolUse`: Before Claude uses any tool * `PostToolUse`: After successful tool execution * `SessionStart`: At beginning of session * `UserPromptSubmit`: When user submits prompt * `SubagentStart`: When sub-agent launches #### Example: Auto-Format Hook ```json theme={null} { "hooks": { "PostToolUse": [ { "matcher": "Write|Edit", "hooks": [ { "type": "command", "command": "${CLAUDE_PLUGIN_ROOT}/scripts/format-excel.sh", "description": "Auto-format Excel files after creation" } ] } ] } } ``` #### Hook Decision Flow Hooks execute shell scripts that return JSON: ```json theme={null} { "decision": "approve", "reason": "File formatted successfully", "systemMessage": "Added conditional formatting and formulas" } ``` *** ### Component 5 — Skills **Purpose**: Instructions Claude reads automatically when relevant to task **File Format**: `SKILL.md` files in subdirectories of `skills/` #### Example: Legal Clause Taxonomy Skill ```markdown theme={null} --- name: legal-clause-taxonomy description: Classification system for contract clauses. Use when analyzing legal documents. --- # Legal Clause Taxonomy When analyzing contracts, identify these clause categories: ## Risk Clauses - **Indemnification**: Who compensates whom for damages (look for "indemnify", "hold harmless") - **Limitation of Liability**: Caps on damages (look for "aggregate liability shall not exceed") - **Warranty Disclaimers**: Limitation of guarantees (look for "AS IS", "WITHOUT WARRANTY") ## Operational Clauses - **Termination Rights**: How contract can be ended (look for "terminate for convenience", notice periods) - **Auto-Renewal**: Automatic contract extension (look for "evergreen", "automatically renew") - **Governing Law**: Which jurisdiction applies (look for "governed by laws of") ## Intellectual Property - **IP Ownership**: Who owns work product (look for "work for hire", "assigns all rights") - **License Grants**: Permission to use IP (look for "non-exclusive license", scope, territory) Flag as **HIGH RISK** if you find: - Unlimited liability - Perpetual licenses with broad scope - One-sided indemnification (we indemnify them, they don't indemnify us) ``` #### Difference from Commands * **Commands**: User manually invokes (`/legal:discover`) * **Skills**: Claude automatically uses when task context matches (e.g., user uploads contract and Legal Taxonomy skill activates) *** ### Plugin Manifest Schema The `plugin.json` file defines plugin metadata. Complete schema: ```json theme={null} { "name": "rfp-assassin", "version": "1.2.0", "description": "Extract 100% of requirements from RFP documents", "author": { "name": "Your Team", "email": "team@company.com", "url": "https://github.com/yourteam" }, "homepage": "https://docs.yourcompany.com/rfp-assassin", "repository": "https://github.com/yourteam/rfp-assassin", "license": "MIT", "keywords": ["rfp", "sales", "requirements", "proposals"], "commands": ["./custom/commands/special.md"], "agents": "./custom/agents/", "skills": "./custom/skills/", "hooks": "./config/hooks.json", "mcpServers": "./mcp-config.json", "lspServers": "./.lsp.json", "dependencies": { "plugins": ["base-analytics"], "mcpServers": ["salesforce"] }, "configuration": { "env": { "DEFAULT_OUTPUT_FORMAT": "excel", "MAX_RFP_PAGES": "200" } } } ``` #### Critical Rules * Manifest location: `.claude-plugin/plugin.json` (MUST be in this directory) * Component directories: `commands/`, `agents/`, `skills/`, `hooks/` MUST be at plugin root (NOT inside `.claude-plugin/`) * Path references: Use `${CLAUDE_PLUGIN_ROOT}` variable for portability * Naming convention: kebab-case for all files and directories *** ## Step-by-Step Setup Guide ### Installing Pre-Built Plugins Anthropic provides 10+ official plugins for common functions: **Official Plugin Library**: * **Productivity**: Task management, calendar, workflows * **Enterprise Search**: Find info across company tools * **Sales**: Prospect research, deal prep * **Finance**: Financial analysis, modeling, metrics * **Data**: Query, visualize, interpret datasets * **Legal**: Document review, risk flagging, compliance * **Marketing**: Content drafting, campaign planning * **Customer Support**: Triage issues, draft responses * **Product Management**: Specs, roadmaps, prioritization * **Biology Research**: Literature search, result analysis #### Installation Process ```text theme={null} 1. Open Claude Desktop → Navigate to Cowork tab 2. Click "Plugins" in left sidebar 3. Browse library → Find plugin you need 4. Click plugin → Review description and components 5. Click "Install" button 6. Plugin downloads and activates automatically ``` #### Alternative: Upload Custom Plugin ```text theme={null} 1. Click "Upload plugin" button 2. Select plugin folder or ZIP file from your computer 3. Claude validates manifest and installs ``` #### Using Installed Plugins * Type `/` or click `+` button to see available commands * Commands appear as `/plugin-name:command-name` * Example: If you installed "sales" plugin, you will see `/sales:research-prospect`, `/sales:prep-discovery`, etc. *** ### Customizing Plugins After installing, tailor plugins to your workflow: ```text theme={null} 1. View installed plugin in Plugins sidebar 2. Click "Customize" button (upper right corner) 3. Claude automatically prompts: "How would you like to customize this plugin?" 4. Describe changes: - "Add integration with our internal CRM at https://crm.company.com" - "Change output format to Google Sheets instead of Excel" - "Include our brand voice guidelines when drafting content" 5. Click "Let's go" → Claude modifies plugin configuration 6. Test customized plugin with sample task ``` #### Example Customization **Default Sales Plugin**: * Uses generic prospect research (LinkedIn, Crunchbase) * Outputs to Excel **After Customization**: * Connects to your Salesforce instance (via MCP) * Pulls existing customer data to identify upsell opportunities * Outputs to Google Sheets shared with sales team * Includes your company's sales methodology (MEDDIC, SPIN, etc.) in Skills *** ### Creating a Simple Plugin from Scratch **Use Case**: Create a "Resume Screener" plugin for HR workflows #### Step 1: Create Plugin Directory ```bash theme={null} mkdir -p resume-screener/.claude-plugin cd resume-screener ``` #### Step 2: Create Manifest Create `.claude-plugin/plugin.json`: ```json theme={null} { "name": "resume-screener", "version": "1.0.0", "description": "Screen and rank job candidates from resume folders", "author": { "name": "Your HR Team" }, "keywords": ["hr", "recruiting", "resumes", "ats"] } ``` #### Step 3: Create Slash Command Create `commands/screen-resumes.md`: ```markdown theme={null} --- description: Screen resumes against job requirements and rank candidates arguments: --- # Resume Screening Workflow You are an expert ATS (Applicant Tracking System) performing initial resume screening. ## Inputs - **Job Description**: $ARGUMENTS (first argument - path to JD file) - **Resume Folder**: $ARGUMENTS (second argument - folder of resumes) ## Process ### Step 1: Extract Job Requirements Read the job description and extract: - **Required Skills**: Must-have technical and soft skills - **Required Experience**: Minimum years in relevant domain - **Required Education**: Degree level and fields - **Nice-to-Have**: Preferred qualifications - **Location**: Remote/hybrid/onsite requirements ### Step 2: Parse Resumes (Parallel Sub-agents) For each resume in the folder: 1. Extract candidate information: - Name, contact info, location - Work experience (titles, companies, dates, achievements) - Skills (technical and soft) - Education (degrees, institutions, years) 2. Calculate match score: - Required skills match: % of must-haves present - Experience: Does candidate meet minimum years? - Education: Does candidate have required degree? - Nice-to-have bonus: Extra points for preferred qualifications 3. Flag red flags: - Employment gaps >12 months - Job hopping (<1 year tenures) - Skill mismatches ### Step 3: Rank and Output Generate Excel spreadsheet with: - **Columns**: Rank, Name, Match %, Years Experience, Key Skills, Education, Location, Interview Recommendation - **Sorting**: Ranked by match score (highest first) - **Conditional Formatting**: - Green: Match >80% (Strong Yes) - Yellow: Match 60-80% (Yes) - Orange: Match 40-60% (Maybe) - Red: Match <40% (No) - **Formulas**: Auto-calculate match percentage based on skills present Save output as: candidate-rankings.xlsx ``` #### Step 4: Create Sub-agent for Resume Parsing Create `agents/resume-parser.md`: ```yaml theme={null} --- name: resume-parser description: Extract structured data from individual resumes model: sonnet-4.5 tools: [Read] permissions: file_access: read_only --- You are a resume parser. Your task is to extract structured information from a single resume and return it in JSON format. Extract: 1. **Contact**: name, email, phone, LinkedIn, location 2. **Experience**: List of jobs with title, company, start_date, end_date, key_achievements 3. **Skills**: Array of technical and soft skills mentioned 4. **Education**: Degrees with institution, field, graduation_year Return only valid JSON, no commentary: { "contact": { "name": "Jane Smith", "email": "jane@email.com", "phone": "555-1234", "linkedin": "linkedin.com/in/janesmith", "location": "San Francisco, CA" }, "experience": [ { "title": "Senior Product Manager", "company": "Tech Corp", "start_date": "2020-01", "end_date": "2025-12", "years": 5, "achievements": ["Launched 3 products", "Grew ARR 200%"] } ], "skills": ["Product Strategy", "SQL", "Figma", "Stakeholder Management"], "education": [ { "degree": "MBA", "institution": "Stanford GSB", "field": "Business Administration", "graduation_year": 2019 } ] } ``` #### Step 5: Create Skills (Optional) Create `skills/ats-criteria/SKILL.md`: ```markdown theme={null} --- name: ats-criteria description: ATS screening best practices and bias mitigation. Use when screening candidates. --- # ATS Screening Criteria ## Best Practices ### Skills Matching - Exact match: "Python" in JD → look for "Python" in resume - Synonym awareness: "JavaScript" = "JS" = "ECMAScript" - Avoid keyword stuffing detection: Flag resumes with >50 skill mentions ### Experience Calculation - Count only relevant experience (not total career length) - Startup experience: Weight 1.5x (broader responsibility) - Internships: Count as 0.5x full-time experience ### Education Evaluation - Bootcamp graduates: Acceptable for technical roles if strong portfolio - Self-taught: Evaluate based on project work, not credentials - Degree field mismatch: Don't auto-reject (career changers can be excellent) ## Bias Mitigation **Remove from consideration** (to avoid discrimination): - Age (graduation dates, years of experience can indicate age) - Gender (inferred from name) - Race/ethnicity - Parental status (resume gaps often indicate caregiving) **Focus on**: - Demonstrated skills (projects, achievements, results) - Growth trajectory (promotions, increasing responsibility) - Cultural adds (diverse perspectives) not "cultural fit" **Red Flag False Positives**: - Employment gaps: May indicate caregiving, health, education → ask, don't auto-reject - Job hopping: Tech industry norm is 2-3 years → recalibrate expectations - Non-traditional paths: Bootcamp grads, self-taught often outperform ## Legal Compliance - EEOC guidelines: No discrimination based on protected classes - Blind screening: Remove names/photos before ranking if possible - Document decisions: Explain why candidates were rejected (defensible criteria) ``` #### Step 6: Test Locally ```bash theme={null} # Navigate to parent directory containing your plugin cd .. # Test plugin before installing claude --plugin-dir ./resume-screener # In Claude session, try your command /resume-screener:screen-resumes job-description.txt /path/to/resumes/ ``` #### Step 7: Install Plugin Once tested, install permanently: **Option A: Install to User Scope** (available across all projects) ```bash theme={null} claude plugin install ./resume-screener --scope user ``` **Option B: Install to Project Scope** (shared with team via git) ```bash theme={null} claude plugin install ./resume-screener --scope project ``` **Option C: Upload via Cowork UI** ```text theme={null} 1. Open Claude Desktop → Cowork tab 2. Click "Plugins" → "Upload plugin" 3. Select resume-screener folder 4. Confirm installation ``` *** ## Use Case 1 — Sales: The "RFP Assassin" **Mission**: Extract 100% of requirements from RFP documents and map to your capabilities ### The Problem Sales teams receive 50-200 page RFPs with requirements scattered across sections. Manual extraction takes 4-8 hours and misses 15-20% of requirements, leading to incomplete proposals and lost deals. ### Plugin Architecture ```text theme={null} Plugin Name: rfp-assassin Components: - Slash Command: /rfp:extract - Sub-agents: - requirement-extractor.md (parallel per section) - criteria-parser.md (evaluation rubrics) - timeline-mapper.md (milestones, deadlines) - Skills: - rfp-taxonomy/ (SOW structures, compliance matrices) - requirement-classification/ (mandatory vs optional logic) - Hook: PostToolUse → auto-format Excel with conditional formatting - MCP: Salesforce (pull existing opportunity data) ``` ### Workflow Execution #### Step 1: User Invokes Command ```text theme={null} /rfp:extract /RFPs/Acme-Corp-RFP-2026/ ``` #### Step 2: Orchestrator Analyzes RFP Structure Claude reads the RFP folder, identifies document sections: * `01-Introduction.pdf` (10 pages) * `02-Technical-Requirements.pdf` (45 pages) * `03-Compliance-Security.pdf` (30 pages) * `04-Pricing-Template.xlsx` * `05-Evaluation-Criteria.pdf` (8 pages) #### Step 3: Parallel Sub-agent Deployment Orchestrator spawns sub-agents: | Sub-agent | Assigned Document | Task | | ------------------------- | ---------------------- | -------------------------------------------------- | | `requirement-extractor-1` | Technical Requirements | Extract all functional/technical requirements | | `requirement-extractor-2` | Compliance-Security | Extract security, privacy, regulatory requirements | | `criteria-parser` | Evaluation Criteria | Parse scoring rubric, weightings, decision factors | | `timeline-mapper` | Introduction | Extract deadlines, milestones, key dates | #### Step 4: Sub-agents Return Findings Each sub-agent returns structured JSON: ```json theme={null} { "section": "Technical Requirements", "requirements": [ { "id": "REQ-001", "category": "Integration", "text": "System must integrate with SAP ERP via REST API", "mandatory": true, "page": 23, "keywords": ["SAP", "REST API", "integration"] }, { "id": "REQ-002", "category": "Security", "text": "All data must be encrypted at rest using AES-256", "mandatory": true, "page": 24, "keywords": ["encryption", "AES-256", "data at rest"] } ] } ``` #### Step 5: Orchestrator Synthesizes Main Claude agent: 1. Merges all requirement lists 2. De-duplicates (same requirement mentioned in multiple sections) 3. Cross-references with your capability database (if MCP connected) 4. Identifies gaps (requirements you can't meet) 5. Flags risks (tight timelines, unusual terms) #### Step 6: Output Generation **Excel Workbook: `Acme-Corp-RFP-Analysis.xlsx`** **Tab 1: Requirements Matrix** | Req ID | Category | Requirement | Mandatory | Page | Our Coverage | Gap/Risk | Owner | | ------- | ----------- | ---------------------------- | --------- | ---- | -------------- | ---------------- | ----------- | | REQ-001 | Integration | SAP ERP via REST API | Yes | 23 | Implemented | None | Engineering | | REQ-002 | Security | AES-256 encryption at rest | Yes | 24 | Implemented | None | Security | | REQ-018 | Compliance | HIPAA BAA required | Yes | 67 | In progress | Confirm timeline | Legal | | REQ-042 | Functional | Mobile app with offline sync | No | 89 | Not roadmapped | Competitive risk | Product | **Tab 2: Evaluation Criteria** | Criterion | Weight | Our Strength | Recommended Emphasis | | ------------- | ------ | ------------ | ---------------------------------- | | Technical Fit | 40% | Strong | Lead with integration capabilities | | Price | 25% | Moderate | Bundle services to increase value | | Experience | 20% | Strong | Highlight similar customer wins | | Support | 15% | Strong | Emphasize 24/7 coverage | **Tab 3: Timeline** | Milestone | Date | Days Until | Status | | ------------- | ------ | ---------- | ----------- | | Questions Due | Feb 15 | 5 days | Pending | | Proposal Due | Mar 1 | 19 days | In Progress | | Presentations | Mar 15 | 33 days | Not Started | | Decision | Apr 1 | 50 days | Not Started | **Tab 4: Risk Summary** | Risk | Severity | Mitigation | | ----------------------------- | -------- | -------------------------------------------- | | HIPAA compliance timing | High | Accelerate BAA process, have legal confirm | | Mobile offline not roadmapped | Medium | Offer custom development or partner solution | | Aggressive timeline | Medium | Pre-stage resources, parallel workstreams | #### Evidence **Time Savings**: Manual RFP analysis takes 4-8 hours. This plugin completes in 15-30 minutes. **Accuracy**: Sub-agents catch requirements buried in appendices that humans miss. One customer found 23 requirements in "Exhibit B" that were not in the main document. *** ## Use Case 2 — Finance: The "10-K Analyst" **Mission**: Audit 5 years of public company financials from SEC filings ### The Problem Financial analysts spend days manually extracting data from 10-K filings (often 200+ pages each). Data is scattered across narrative sections, footnotes, and exhibits. Year-over-year comparisons require tedious spreadsheet work. ### Plugin Architecture ```text theme={null} Plugin Name: 10k-analyst Components: - Slash Command: /finance:analyze-10k - Sub-agents (one per fiscal year): - 10k-extractor-FY2021.md - 10k-extractor-FY2022.md - 10k-extractor-FY2023.md - 10k-extractor-FY2024.md - 10k-extractor-FY2025.md - Skills: - gaap-standards/ (accounting rule interpretations) - financial-ratios/ (calculation formulas) - sec-filing-structure/ (where to find specific data) - MCP: SEC EDGAR API (fetch filings automatically) ``` ### Workflow Execution #### Step 1: User Invokes Command ```text theme={null} /finance:analyze-10k TSLA 2021-2025 ``` #### Step 2: Fetch 10-K Filings Claude (via SEC EDGAR MCP): 1. Queries SEC for Tesla 10-K filings (2021-2025) 2. Downloads all 5 annual reports 3. Identifies key sections (Item 7: MD\&A, Item 8: Financial Statements) #### Step 3: Parallel Sub-agent Analysis 5 sub-agents process simultaneously, each handling one fiscal year: ```text theme={null} Sub-agent: 10k-extractor-FY2023 Task: Extract all financial data from Tesla FY2023 10-K Extract: 1. Income Statement (Revenue, COGS, Operating Expenses, Net Income) 2. Balance Sheet (Assets, Liabilities, Equity) 3. Cash Flow (Operating, Investing, Financing) 4. Key Metrics (mentioned in MD&A) 5. Risk Factors (new risks added this year) 6. Segment Data (Automotive, Energy, Services) Return structured JSON with exact figures and page citations. ``` #### Step 4: Orchestrator Synthesis Main Claude agent: 1. Validates data consistency (e.g., ending cash FY2022 = starting cash FY2023) 2. Calculates derived metrics (margins, ratios, growth rates) 3. Identifies trends and anomalies 4. Extracts narrative insights from MD\&A sections 5. Compares against industry benchmarks (if available) #### Step 5: Output Generation **Excel Workbook: `TSLA-5-Year-Analysis.xlsx`** **Tab 1: Income Statement (with YoY formulas)** | Line Item | FY2021 | FY2022 | FY2023 | FY2024 | FY2025 | 5Y CAGR | | ---------------- | ------- | ------- | ------- | -------- | -------- | ------- | | Revenue | \$53.8B | \$81.5B | \$96.8B | \$112.4B | \$128.7B | =CAGR | | Gross Profit | \$13.6B | \$20.9B | \$17.7B | \$22.1B | \$26.8B | =CAGR | | Operating Income | \$6.5B | \$13.7B | \$8.9B | \$12.3B | \$15.1B | =CAGR | | Net Income | \$5.5B | \$12.6B | \$15.0B | \$14.8B | \$17.2B | =CAGR | **Tab 2: Balance Sheet Evolution** | Category | FY2021 | FY2022 | FY2023 | FY2024 | FY2025 | Trend | | -------------------- | ------- | ------- | -------- | -------- | -------- | ----- | | Cash and Equivalents | \$17.6B | \$22.2B | \$29.1B | \$33.6B | \$36.5B | Up | | Total Assets | \$62.1B | \$82.3B | \$106.6B | \$128.4B | \$145.2B | Up | | Total Debt | \$6.8B | \$5.7B | \$5.2B | \$7.1B | \$8.3B | Flat | | Shareholders' Equity | \$30.2B | \$44.7B | \$62.6B | \$78.3B | \$91.5B | Up | **Tab 3: Cash Flow Analysis** | Category | FY2021 | FY2022 | FY2023 | FY2024 | FY2025 | | ------------------- | -------- | -------- | -------- | --------- | --------- | | Operating Cash Flow | \$11.5B | \$14.7B | \$13.3B | \$15.8B | \$18.2B | | CapEx | (\$6.5B) | (\$7.2B) | (\$8.9B) | (\$10.1B) | (\$11.3B) | | Free Cash Flow | \$5.0B | \$7.5B | \$4.4B | \$5.7B | \$6.9B | | FCF Margin | 9.3% | 9.2% | 4.5% | 5.1% | 5.4% | **Tab 4: Ratio Dashboard (with conditional formatting)** | Ratio | FY2021 | FY2022 | FY2023 | FY2024 | FY2025 | Industry Avg | | ---------------- | ------ | ------ | ------ | ------ | ------ | ------------ | | Gross Margin | 25.3% | 25.6% | 18.3% | 19.7% | 20.8% | 18.5% | | Operating Margin | 12.1% | 16.8% | 9.2% | 10.9% | 11.7% | 8.2% | | ROE | 18.2% | 28.2% | 24.0% | 18.9% | 18.8% | 12.5% | | Current Ratio | 1.4x | 1.5x | 1.7x | 1.8x | 1.9x | 1.2x | | Debt/Equity | 0.23x | 0.13x | 0.08x | 0.09x | 0.09x | 0.45x | **Tab 5: Narrative Insights (from MD\&A)** ```text theme={null} KEY THEMES FROM MANAGEMENT DISCUSSION: FY2023 Margin Compression: - Management cited "aggressive pricing strategy" to maintain market share - Raw material costs (lithium, nickel) increased 15% YoY - New factory ramp costs in Austin and Berlin FY2024-2025 Recovery: - Pricing stabilized; focus shifted to cost reduction - Manufacturing efficiency improved (vehicles per employee up 12%) - Energy Storage segment growing faster than Automotive (45% vs 15%) Risk Factors Evolution: - FY2023: Added "increased competition from legacy automakers" - FY2024: Added "regulatory uncertainty on autonomous driving" - FY2025: Added "supply chain concentration in China" ``` *** ## Use Case 3 — Legal: The "Discovery Drone" **Mission**: Find risk clauses across 50+ contracts in minutes ### The Problem Legal teams inherit contracts from acquisitions, vendor changes, or simply poor organization. Finding specific clause types (indemnification, liability caps, auto-renewal) across dozens of documents takes weeks of manual review. ### Plugin Architecture ```text theme={null} Plugin Name: discovery-drone Components: - Slash Command: /legal:discover - Sub-agents (specialized by contract type): - contract-analyzer-msa.md (Master Service Agreements) - contract-analyzer-nda.md (Non-Disclosure Agreements) - contract-analyzer-sow.md (Statements of Work) - contract-analyzer-saas.md (SaaS/Subscription Agreements) - contract-analyzer-license.md (Software Licenses) - Skills: - legal-clause-taxonomy/ (clause identification patterns) - risk-scoring/ (severity classification logic) - jurisdiction-rules/ (state/country-specific considerations) ``` ### Workflow Execution #### Step 1: User Invokes Command ```text theme={null} /legal:discover indemnification-liability /contracts/vendor-agreements/ ``` #### Step 2: Contract Inventory Claude scans the folder: * 12 MSAs (Master Service Agreements) * 8 SOWs (Statements of Work) * 15 NDAs (Non-Disclosure Agreements) * 10 SaaS Agreements * 5 License Agreements * **Total: 50 contracts** #### Step 3: Parallel Sub-agent Deployment Sub-agents process by contract type (leveraging specialized prompts): | Sub-agent | Contracts | Focus | | ---------------- | ---------- | ---------------------------------------------- | | MSA Analyzer | 12 MSAs | Indemnification, liability caps, termination | | SOW Analyzer | 8 SOWs | Scope creep risks, change order terms | | NDA Analyzer | 15 NDAs | Definition of confidential info, term duration | | SaaS Analyzer | 10 SaaS | Data ownership, SLA penalties, auto-renewal | | License Analyzer | 5 Licenses | IP scope, audit rights, termination triggers | #### Step 4: Clause Extraction Each sub-agent extracts: ```json theme={null} { "contract": "Acme-Corp-MSA-2023.pdf", "clauses": [ { "type": "indemnification", "party_indemnifying": "Us", "party_indemnified": "Acme Corp", "scope": "IP infringement, third-party claims, data breaches", "cap": "Unlimited", "page": 15, "exact_text": "Vendor shall indemnify, defend, and hold harmless Client from any and all claims...", "risk_score": "HIGH", "risk_reason": "Unlimited indemnification, one-sided (we indemnify them, they don't indemnify us)" }, { "type": "liability_limitation", "our_cap": "12 months fees", "their_cap": "None specified", "consequential_damages": "Excluded for us, not for them", "page": 16, "risk_score": "HIGH", "risk_reason": "Asymmetric liability caps favor counterparty" } ] } ``` #### Step 5: Risk Aggregation Orchestrator consolidates findings: **Risk Scoring Logic:** * **HIGH**: Unlimited liability, one-sided indemnification, perpetual IP grants * **MEDIUM**: Short notice periods (under 30 days), unfavorable auto-renewal, broad audit rights * **LOW**: Standard market terms, balanced risk allocation #### Step 6: Output Generation **Excel Workbook: `Contract-Risk-Analysis.xlsx`** **Tab 1: Executive Summary** | Risk Level | Count | % of Portfolio | Action Required | | ---------- | ----- | -------------- | ----------------------- | | HIGH | 8 | 16% | Immediate renegotiation | | MEDIUM | 15 | 30% | Review at renewal | | LOW | 27 | 54% | Standard monitoring | **Tab 2: High-Risk Contracts** | Contract | Counterparty | Risk Type | Issue | Page | Recommendation | | ---------------- | ------------ | --------------- | -------------------------- | ---- | ------------------------------ | | MSA-2023-001 | Acme Corp | Indemnification | Unlimited, one-sided | 15 | Negotiate cap, mutual terms | | MSA-2023-001 | Acme Corp | Liability | Asymmetric caps | 16 | Add our liability cap | | SaaS-2022-007 | DataCo | Auto-renewal | 90-day notice, 3-year term | 22 | Calendar reminder, renegotiate | | License-2021-003 | SoftwareInc | IP Rights | Perpetual, broad scope | 8 | Limit scope to specific use | **Tab 3: Indemnification Analysis** | Contract | We Indemnify Them | They Indemnify Us | Mutual | Our Cap | Their Cap | Risk | | ---------- | ----------------- | --------------------- | ------ | ---------- | ---------- | ---- | | Acme MSA | Yes (broad) | No | No | Unlimited | N/A | HIGH | | Beta SOW | Yes (IP only) | Yes (IP only) | Yes | 12 mo fees | 12 mo fees | LOW | | Gamma SaaS | Yes (data breach) | Yes (service failure) | Yes | \$1M | \$1M | LOW | **Tab 4: Auto-Renewal Tracker** | Contract | Current Term End | Auto-Renews? | Renewal Term | Notice Required | Notice Deadline | Status | | ----------- | ---------------- | ------------ | ------------ | --------------- | --------------- | ------------- | | DataCo SaaS | Dec 31, 2026 | Yes | 3 years | 90 days | Oct 2, 2026 | Action needed | | CloudHost | Mar 15, 2026 | Yes | 1 year | 30 days | Feb 13, 2026 | Past deadline | | ToolVendor | Jun 30, 2026 | No | - | - | - | No action | *** ## Use Case 4 — Product: The "Voice of Customer" Engine **Mission**: Cluster 1,000s of support tickets into actionable product themes ### The Problem Product managers drown in unstructured customer feedback. Support tickets, NPS comments, and feature requests pile up. Manually reading thousands of tickets is impossible; sampling misses patterns. ### Plugin Architecture ```text theme={null} Plugin Name: voc-engine Components: - Slash Command: /product:voc-cluster - Sub-agents: - ticket-themer.md (batch extraction via API) - cluster-engine.md (pattern synthesis) - impact-scorer.md (prioritization by revenue/churn risk) - Skills: - product-taxonomy/ (feature categories, user personas) - sentiment-analysis/ (frustration indicators) - MCP: Zendesk, Intercom, or Jira (live ticket access) ``` ### Two-Stage Pipeline **Why Two Stages?** Processing 5,000 tickets individually in Cowork would be slow and expensive. Instead: 1. **Stage 1**: Use Anthropic's Batch API for high-volume extraction (50% cheaper, async) 2. **Stage 2**: Use Cowork to synthesize, cluster, and prioritize findings #### Stage 1: Batch API Extraction ```python theme={null} import anthropic import json client = anthropic.Anthropic() # Load tickets from export with open('zendesk-export.json') as f: tickets = json.load(f) # Create batch request batch = client.messages.batches.create( requests=[ { "custom_id": f"ticket-{ticket['id']}", "params": { "model": "claude-sonnet-4-5-20250929", "max_tokens": 512, "messages": [{ "role": "user", "content": f"""Analyze this support ticket. Ticket #{ticket['id']}: Subject: {ticket['subject']} Body: {ticket['description']} Return JSON only: {{ "theme": "one of: mobile, web_ux, integrations, billing, performance, onboarding, reporting, api, other", "sentiment": "satisfied | neutral | frustrated | angry", "pain_point": "1-sentence description of core issue", "feature_request": "specific feature if mentioned, else null", "churn_risk": "low | medium | high" }} """ }] } } for ticket in tickets ] ) print(f"Batch ID: {batch.id}") print(f"Status: {batch.processing_status}") # Poll for completion, then retrieve results ``` **Batch Processing Stats:** * 5,000 tickets processed * Approximately 2 hours async processing * 50% cost reduction vs. synchronous API * Results saved to `ticket-themes.json` #### Stage 2: Cowork Clustering ```text theme={null} /product:voc-cluster ticket-themes.json ``` Claude in Cowork: 1. Loads 5,000 pre-extracted ticket summaries 2. Groups by theme (mobile: 800, integrations: 1,200, etc.) 3. Within each theme, clusters by specific pain point 4. Ranks clusters by frequency x sentiment severity x churn risk 5. Generates actionable recommendations #### Output Generation **Excel Workbook: `VOC-Analysis-Q4-2025.xlsx`** **Tab 1: Theme Distribution** | Theme | Ticket Count | % of Total | Avg Sentiment | High Churn Risk | Priority Score | | ------------ | ------------ | ---------- | ------------- | --------------- | -------------- | | Integrations | 1,247 | 25% | Frustrated | 312 (25%) | 94 | | Mobile | 892 | 18% | Frustrated | 267 (30%) | 89 | | Performance | 756 | 15% | Angry | 189 (25%) | 85 | | Billing | 623 | 12% | Neutral | 62 (10%) | 45 | | Onboarding | 512 | 10% | Neutral | 51 (10%) | 40 | **Tab 2: Top Pain Points (Ranked)** | Rank | Theme | Pain Point | Tickets | Recommendation | | ---- | ------------ | -------------------------------- | ------- | -------------------------------------- | | 1 | Integrations | Salesforce sync breaks daily | 423 | Rebuild sync with retry logic | | 2 | Mobile | Can't approve requests on mobile | 356 | Add approval workflow to mobile app | | 3 | Performance | Dashboard takes 30+ seconds | 289 | Optimize database queries, add caching | | 4 | Integrations | No Slack notifications | 267 | Build Slack integration | | 5 | Mobile | App crashes on Android | 234 | Investigate Android-specific bugs | **Tab 3: Feature Requests (Extracted)** | Feature | Request Count | Theme | Competitive Pressure | Revenue at Risk | | ----------------------- | ------------- | ------------ | -------------------------- | --------------- | | Slack integration | 267 | Integrations | High (competitors have it) | \$450K ARR | | Mobile approvals | 356 | Mobile | Medium | \$380K ARR | | Custom dashboards | 198 | Reporting | Low | \$220K ARR | | API rate limit increase | 145 | API | Medium | \$180K ARR | **Tab 4: Churn Risk Cohort** | Customer | Plan | MRR | Tickets (90d) | Avg Sentiment | Primary Complaint | Risk Level | | --------- | ---------- | ------ | ------------- | ------------- | --------------------- | ---------- | | Acme Corp | Enterprise | \$15K | 23 | Angry | Salesforce sync | Critical | | Beta Inc | Pro | \$2.5K | 18 | Frustrated | Mobile crashes | Critical | | Gamma LLC | Enterprise | \$12K | 15 | Frustrated | Dashboard performance | High | *** ## Use Case 5 — Marketing: The "Voice DNA" Extractor **Mission**: Extract brand voice patterns from your content corpus to create a style guide ### The Problem Brands struggle to maintain consistent voice across teams, agencies, and AI tools. Existing style guides are vague ("be professional but friendly"). New writers and AI assistants produce off-brand content. ### Plugin Architecture ```text theme={null} Plugin Name: voice-dna Components: - Slash Command: /marketing:extract-voice - Sub-agents: - tone-analyzer.md (emotional register, formality level) - vocabulary-mapper.md (word choices, phrases to use/avoid) - pattern-extractor.md (sentence structures, paragraph patterns) - example-curator.md (best examples per category) - Skills: - linguistics-frameworks/ (tone dimensions, readability metrics) - brand-voice-templates/ (B2B, B2C, technical, casual) ``` ### Workflow Execution #### Step 1: User Invokes Command ```text theme={null} /marketing:extract-voice /content/approved-2024/ ``` #### Step 2: Content Inventory Claude scans the folder: * 45 blog posts * 12 case studies * 8 white papers * 200 social media posts * 30 email campaigns * 15 product pages * **Total: 310 content pieces** #### Step 3: Parallel Analysis | Sub-agent | Focus | Extracts | | ----------------- | ------------------ | -------------------------------------------------- | | Tone Analyzer | Emotional register | Confidence level, warmth, formality, humor usage | | Vocabulary Mapper | Word choices | Preferred terms, avoided terms, jargon handling | | Pattern Extractor | Structure | Sentence length, paragraph structure, CTA patterns | | Example Curator | Best samples | Top examples of each voice element | #### Step 4: Output Generation **Document: `Brand-Voice-DNA-Guide.docx`** **Executive Summary** Based on analysis of 310 approved content pieces, your brand voice can be characterized as: **"Approachable Authority"** — Expert knowledge delivered with warmth and clarity, avoiding both stiff formality and excessive casualness. **Core Voice Pillars** **Pillar 1: Confident but Not Arrogant** | DO | DON'T | | ----------------------------------------------- | ---------------------------------- | | "Our research shows..." | "Obviously, anyone would know\..." | | "We've found that customers see 3x improvement" | "We're the best in the industry" | | "Here's what we recommend" | "You must do this" | **Pillar 2: Clear Over Clever** | DO | DON'T | | ---------------------------------- | ----------------------------------------------------------- | | "This feature helps you save time" | "Revolutionize your temporal efficiency" | | "Connect your CRM in 3 clicks" | "Seamlessly integrate your customer relationship ecosystem" | | Explain jargon on first use | Assume reader knows acronyms | **Pillar 3: Empathetic Problem-Solver** | DO | DON'T | | ------------------------------------------- | ----------------------------- | | "We know onboarding can be overwhelming" | "Onboarding is simple" | | "Many teams struggle with..." | "If you're having trouble..." | | Acknowledge pain before presenting solution | Jump straight to features | **Vocabulary Signatures** Preferred Terms: | Instead of | Use | | ------------- | -------------------------- | | Utilize | Use | | Leverage | Use, apply | | Synergy | Collaboration, teamwork | | Disrupt | Change, improve, transform | | Revolutionary | Significant, meaningful | | Cutting-edge | Modern, current, new | **Technical Jargon Handling** Rule: Define on first use, then use freely. **Emotional Vocabulary** | Emotion | Our Words | Avoid | | ---------- | --------------------------------- | --------------------------------- | | Excitement | Excited, thrilled, can't wait | Stoked, pumped, psyched | | Confidence | Confident, certain, proven | Guaranteed, foolproof, unbeatable | | Empathy | Understand, recognize, appreciate | Sorry (overused), unfortunately | **Sentence and Paragraph Patterns** Sentence Length: * **Target**: 15-20 words average * **Mix**: Alternate short (8-12) and medium (18-25) sentences * **Avoid**: Sentences over 30 words Paragraph Structure: * **Blog posts**: 2-4 sentences per paragraph * **Product pages**: 1-2 sentences per paragraph * **White papers**: 3-5 sentences per paragraph **AI Prompt Snippet** Use this when prompting AI tools to write in your voice: ```text theme={null} Write in [Company]'s brand voice: - Tone: Confident but approachable, expert but not arrogant - Avoid: Jargon without explanation, superlatives (revolutionary, game-changing), passive voice - Prefer: Short sentences (15-20 words), concrete examples, "you" over "users" - Structure: Lead with benefit, explain how, end with CTA - Sample: "Our research shows that teams using automated workflows save 10 hours per week. Here's how to set one up in 5 minutes." ``` *** ## Use Case 6 — HR: The "Resume Radar" **Mission**: Screen hundreds of resumes and rank candidates against job requirements ### The Problem HR teams receive 200-500 resumes per open role. Manual screening takes 30+ hours and introduces inconsistency. Good candidates get buried; bias affects decisions. ### Plugin Architecture ```text theme={null} Plugin Name: resume-radar Components: - Slash Command: /hr:screen-resumes - Sub-agents: - resume-parser.md (parallel, one per resume) - skills-matcher.md (compare to requirements) - red-flag-detector.md (gaps, inconsistencies) - Skills: - ats-criteria/ (matching algorithms, scoring weights) - bias-mitigation/ (blind screening guidance) - skill-taxonomy/ (synonym matching, equivalent skills) ``` ### Workflow Execution #### Step 1: User Invokes Command ```text theme={null} /hr:screen-resumes senior-pm-job-description.pdf /candidates/PM-role-2026/ ``` #### Step 2: Requirements Extraction Claude parses job description: ```text theme={null} Extracted Requirements: REQUIRED SKILLS (Must-Have): - Product management: 5+ years - B2B SaaS experience - SQL proficiency - Roadmap ownership - Stakeholder management - Agile/Scrum certification or experience REQUIRED EDUCATION: - Bachelor's degree (any field) - MBA preferred but not required NICE-TO-HAVE: - Fintech industry experience - Technical background (engineering degree or coding experience) - Public speaking / thought leadership - Experience with AI/ML products LOCATION: - Remote OK (US timezone overlap required) ``` #### Step 3: Parallel Resume Processing 200 sub-agents process simultaneously: ```json theme={null} { "name": "Jane Smith", "email": "jane@email.com", "location": "San Francisco, CA", "linkedin": "linkedin.com/in/janesmith", "experience": [ { "title": "Senior Product Manager", "company": "Stripe", "duration": "2021-2025 (4 years)", "achievements": [ "Launched 3 payment products generating $50M ARR", "Led team of 8 engineers, 2 designers", "Reduced checkout abandonment 23%" ] }, { "title": "Product Manager", "company": "Plaid", "duration": "2018-2021 (3 years)", "achievements": [ "Built bank connection API used by 2000+ apps", "Managed $10M product budget" ] } ], "skills": ["Product Strategy", "SQL", "Figma", "Jira", "Amplitude", "Python"], "education": [ {"degree": "MBA", "school": "Stanford GSB", "year": 2018}, {"degree": "BS Computer Science", "school": "MIT", "year": 2014} ], "certifications": ["Certified Scrum Product Owner (CSPO)"] } ``` #### Step 4: Scoring Algorithm ```text theme={null} Match Score = ( Required Skills Match x 40% + Experience Threshold x 30% + Education Fit x 15% + Nice-to-Have Bonus x 15% ) Jane Smith Calculation: - Required Skills: 6/6 = 100% x 0.40 = 40 points - Experience: 7 years (exceeds 5+) = 100% x 0.30 = 30 points - Education: MBA + BS CS = 100% x 0.15 = 15 points - Nice-to-Have: Fintech (yes), Technical (yes), AI (no) = 67% x 0.15 = 10 points TOTAL: 95 points → Strong Yes ``` #### Step 5: Output Generation **Excel Workbook: `PM-Candidates-Ranked.xlsx`** **Tab 1: Candidate Rankings** | Rank | Name | Match % | Years Exp | Required Skills | Nice-to-Have | Red Flags | Recommendation | | ---- | ------------- | ------- | --------- | --------------- | ------------------ | ------------------- | -------------- | | 1 | Jane Smith | 95% | 7 | 6/6 | Fintech, Technical | None | Strong Yes | | 2 | John Chen | 92% | 8 | 6/6 | Technical, AI/ML | None | Strong Yes | | 3 | Sarah Johnson | 88% | 6 | 5/6 | Fintech | Missing SQL | Yes | | 4 | Mike Williams | 82% | 5 | 5/6 | None | None | Yes | | 5 | Emily Brown | 78% | 4 | 5/6 | Technical | Below exp threshold | Phone Screen | | 45 | Alex Turner | 45% | 2 | 2/6 | None | Low experience | No | **Tab 2: Skills Gap Analysis** | Skill | Candidates With | Candidates Without | Gap Severity | | ---------------- | --------------- | ------------------ | ------------ | | SQL | 156 (78%) | 44 (22%) | Low | | Agile/Scrum | 134 (67%) | 66 (33%) | Medium | | B2B SaaS | 112 (56%) | 88 (44%) | High | | Stakeholder Mgmt | 178 (89%) | 22 (11%) | Low | **Tab 3: Diversity Metrics (Anonymized)** | Category | Pool Composition | Advancing (Top 20) | Notes | | -------------------- | ---------------- | ------------------ | ----------------------- | | Technical Background | 45% of pool | 60% advancing | Positive signal | | Non-Traditional Path | 22% of pool | 25% advancing | Inclusive screening | | Career Changers | 15% of pool | 10% advancing | Review borderline cases | *** ## Use Case 7 — Strategy: The "Board Whisperer" **Mission**: Turn scattered notes into a board-ready presentation ### The Problem Executives accumulate strategy notes across documents, emails, Slack threads, and meeting notes. Synthesizing into a coherent board deck takes days. The result often lacks narrative flow. ### Plugin Architecture ```text theme={null} Plugin Name: board-whisperer Components: - Slash Command: /strategy:board-deck - Sub-agents: - metrics-compiler.md (pull KPIs from multiple sources) - narrative-builder.md (synthesize storyline) - risk-identifier.md (extract concerns, mitigations) - comparison-engine.md (vs. plan, vs. prior period) - Skills: - board-deck-templates/ (SaaS metrics, growth-stage, enterprise) - executive-communication/ (pyramid principle, MECE) - financial-modeling/ (ARR calculations, cohort analysis) - MCP: Google Slides (direct deck creation) ``` ### Workflow Execution #### Step 1: User Invokes Command ```text theme={null} /strategy:board-deck Q4-2025 /board-prep/Q4-2025/ ``` #### Step 2: Source Material Inventory Claude scans the folder: * `Q4-financials-draft.xlsx` (CFO's numbers) * `sales-pipeline-update.docx` (CRO's commentary) * `product-roadmap-Q4.pdf` (CPO's update) * `customer-health-scores.csv` (CS team data) * `competitive-intel-notes.txt` (Strategy team) * `CEO-board-notes.md` (bullet points from CEO) * `slack-export-exec-channel.json` (relevant threads) #### Step 3: Parallel Extraction | Sub-agent | Sources | Extracts | | ----------------- | ----------------------------------- | ------------------------------- | | Metrics Compiler | Financials, Pipeline, Health Scores | ARR, NRR, Pipeline, Churn | | Narrative Builder | CEO Notes, Slack | Key messages, storyline | | Risk Identifier | All sources | Concerns mentioned, mitigations | | Comparison Engine | Financials | Actual vs. Plan, QoQ, YoY | #### Step 4: Output Generation **PowerPoint Deck: `Q4-2025-Board-Deck.pptx`** **Slide 1: Executive Summary** Q4 2025: Revenue Growth Accelerating, Unit Economics Improving 3 Key Messages for the Board: 1. **ARR hit \$42M** (+35% YoY), enterprise segment now 40% of new bookings 2. **Unit economics at all-time best**: LTV:CAC ratio improved to 3.6x (from 3.2x) 3. **Recommendation**: Double down on enterprise GTM; request \$5M incremental investment **Slide 2: Financial Performance** | Metric | Q4 Actual | Q4 Plan | Variance | QoQ | YoY | | ---------------- | --------- | ------- | -------- | ---- | ---- | | ARR | \$42.0M | \$45.0M | -7% | +9% | +35% | | Net New ARR | \$3.5M | \$4.0M | -13% | +5% | +28% | | Gross Margin | 78% | 75% | +3pp | +1pp | +4pp | | Operating Margin | -12% | -15% | +3pp | +2pp | +8pp | **Commentary**: Slight miss on ARR driven by 2 enterprise deals slipping to Q1 (signed Jan 5). Margin improvement ahead of plan due to infrastructure optimization. **Slide 3: ARR Bridge** ```text theme={null} Q3 Ending ARR: $38.5M + New Business: +$2.8M + Expansion: +$1.4M - Churn: -$0.5M - Contraction: -$0.2M = Q4 Ending ARR: $42.0M Net Revenue Retention: 118% Gross Revenue Retention: 94% Logo Retention: 92% ``` **Slide 4: Sales Pipeline** | Stage | Deals | Weighted Value | Expected Close | | ------------------ | ------ | -------------- | -------------- | | Discovery | 45 | \$2.3M | Q2 2026 | | Evaluation | 28 | \$4.1M | Q1 2026 | | Negotiation | 12 | \$2.8M | Q1 2026 | | Verbal Commit | 5 | \$1.2M | Jan 2026 | | **Total Pipeline** | **90** | **\$10.4M** | - | **Coverage Ratio**: 2.6x (Target: 3.0x) — Need more top-of-funnel **Slide 5: Product and Roadmap** Shipped in Q4: * Enterprise SSO (unblocked 3 deals worth \$800K) * Advanced Analytics Dashboard (top feature request) * Salesforce Integration v2 (resolved sync reliability issues) Q1 2026 Roadmap: * Mobile App v2 (approval workflows) * AI-powered insights (beta) * SOC 2 Type II certification (enterprise blocker) **Slide 6: Risks and Mitigations** | Risk | Severity | Mitigation | Owner | Status | | ------------------------------------------ | -------- | ------------------------------------------ | ----- | ----------- | | Enterprise concentration: Top 10 = 35% ARR | Medium | Accelerate mid-market motion | CRO | In progress | | Key engineer attrition (2 departures) | High | Counter-offers, backfill recruiting | CTO | Mitigated | | Competitor launched similar feature | Medium | Accelerate AI roadmap, differentiate on UX | CPO | Monitoring | | Sales cycle lengthening (90 to 120 days) | Medium | Add sales engineers, improve demo | CRO | In progress | **Slide 7: The Ask** Board Approval Requested: 1. **\$5M incremental investment** in enterprise sales (4 AEs, 2 SEs, 1 SA) * Expected ROI: \$8M incremental ARR by end of 2026 * Payback: 18 months 2. **Stock option refresh pool** (500K shares) for retention * Competitive pressure from well-funded startups * 3 key engineers received outside offers in Q4 3. **M\&A exploration authorization** for complementary analytics startup * Target identified, early conversations * Potential acqui-hire of 8-person team *** ## Use Case 8 — Operations: The "Process Surgeon" **Mission**: Find bottlenecks across SOPs and recommend optimizations ### The Problem Operations teams maintain dozens of SOPs (Standard Operating Procedures) that evolve independently. Bottlenecks hide in handoffs, approval gates, and manual steps. Process mining tools are expensive and complex. ### Plugin Architecture ```text theme={null} Plugin Name: process-surgeon Components: - Slash Command: /ops:audit-process - Sub-agents: - process-mapper.md (extract steps, owners, durations) - bottleneck-detector.md (identify delays, waste) - dependency-analyzer.md (find blocking relationships) - benchmark-comparator.md (vs. industry standards) - Skills: - lean-six-sigma/ (waste categories, improvement frameworks) - process-mining/ (cycle time analysis, wait time identification) - automation-patterns/ (RPA opportunities, integration points) ``` ### Workflow Execution #### Step 1: User Invokes Command ```text theme={null} /ops:audit-process /sops/customer-onboarding/ ``` #### Step 2: SOP Inventory Claude scans the folder: * `1-sales-handoff.docx` * `2-legal-review.docx` * `3-security-assessment.docx` * `4-technical-setup.docx` * `5-training-scheduling.docx` * `6-go-live-checklist.docx` * `metrics-dashboard.xlsx` #### Step 3: Process Mapping Claude extracts and visualizes: ```text theme={null} CURRENT STATE PROCESS MAP: Step 1: Sales Handoff (Day 1-2) → Sales sends intro email to CS → CS reviews deal notes in Salesforce → CS schedules kickoff call Duration: 2 days | Owner: Sales → CS Step 2: Kickoff Call (Day 3) → Intro meeting with customer → Gather requirements → Set expectations Duration: 1 day | Owner: CS Step 3: Legal Review (Day 4-10) → Customer sends DPA/security questionnaire → Legal reviews and responds → Back-and-forth negotiations Duration: 7 days | Owner: Legal BOTTLENECK IDENTIFIED Step 4: Security Assessment (Day 11-17) → Customer IT sends security checklist → Our security team responds → Remediation if needed Duration: 7 days | Owner: Security BOTTLENECK IDENTIFIED Step 5: Technical Setup (Day 18-22) → Provision tenant → Configure SSO → Data migration Duration: 5 days | Owner: Implementation Step 6: Training (Day 23-25) → Schedule training sessions → Conduct admin training → Conduct end-user training Duration: 3 days | Owner: CS Step 7: Go-Live (Day 26) → Cutover to production → Monitor for issues Duration: 1 day | Owner: CS + Support TOTAL CYCLE TIME: 26 days ``` #### Step 4: Bottleneck Analysis **Excel Workbook: `Onboarding-Process-Audit.xlsx`** **Tab 1: Bottleneck Summary** | Step | Description | Duration | % of Cycle | Type | Recommendation | Savings | | ---- | ------------------- | -------- | ---------- | ------------ | ------------------------------------------------------- | ---------- | | 3 | Legal Review | 7 days | 27% | Manual gate | Pre-approved DPA template, self-serve portal | **5 days** | | 4 | Security Assessment | 7 days | 27% | Manual gate | Published security whitepaper, pre-filled questionnaire | **4 days** | | 1 | Sales Handoff | 2 days | 8% | Coordination | Automated handoff trigger in Salesforce | **1 day** | | 5 | Technical Setup | 5 days | 19% | Manual task | Self-serve provisioning, automated SSO config | **2 days** | **Tab 2: Waste Analysis (Lean Categories)** | Waste Type | Examples Found | Impact | Recommendation | | ------------------- | --------------------------------------------- | ------------- | -------------------------------------------------- | | **Waiting** | Legal queue (avg 3 days before review starts) | 3 days | Add legal headcount or pre-approve templates | | **Defects** | 40% of security questionnaires have errors | 2 days rework | Pre-fill with known answers | | **Over-processing** | Custom DPA negotiation for standard deals | 2 days | Tier customers: standard DPA for deals under \$50K | | **Motion** | CS manually copies data between systems | 1 day | Integrate Salesforce with onboarding tool | **Tab 3: Optimized Process** | Step | Current | Optimized | Savings | How | | ------------------- | ----------- | ----------- | ----------------- | --------------------------------------------- | | Sales Handoff | 2 days | 1 day | 1 day | Auto-trigger on deal close | | Legal Review | 7 days | 2 days | 5 days | Pre-approved DPA, self-serve portal | | Security Assessment | 7 days | 3 days | 4 days | Published security pack, pre-filled responses | | Technical Setup | 5 days | 3 days | 2 days | Self-serve provisioning | | Training | 3 days | 3 days | 0 days | Keep as-is (value-add) | | Go-Live | 1 day | 1 day | 0 days | Keep as-is | | **TOTAL** | **26 days** | **14 days** | **12 days (46%)** | - | *** ## Use Case 9 — Support: The "Escalation Tamer" **Mission**: Summarize angry customer threads and draft de-escalation responses ### The Problem Escalated tickets are emotionally charged and context-heavy. Agents spend 20-30 minutes reading thread history before responding. Poor responses make things worse. ### Plugin Architecture ```text theme={null} Plugin Name: escalation-tamer Components: - Slash Command: /support:deescalate - Sub-agents: - thread-summarizer.md (extract key events, timeline) - sentiment-tracker.md (emotion trajectory) - issue-extractor.md (root cause, customer ask) - response-drafter.md (empathetic reply) - Skills: - crisis-communication/ (de-escalation frameworks) - empathy-frameworks/ (acknowledgment patterns) - compensation-guidelines/ (when/what to offer) - MCP: Zendesk, Intercom, Freshdesk (fetch ticket data) ``` ### Workflow Execution #### Step 1: User Invokes Command ```text theme={null} /support:deescalate TICKET-5678 ``` #### Step 2: Thread Retrieval Claude (via Zendesk MCP) fetches: * 23 messages over 12 days * 5 different agents involved * Customer escalated to VP Support on day 10 #### Step 3: Analysis Output **Document: `TICKET-5678-Analysis.docx`** **Escalation Summary** **Customer**: Acme Corp (Enterprise, \$180K ARR) **Contact**: John Smith, VP Operations **Severity**: Critical (executive escalation) **Duration**: 12 days unresolved **Sentiment Trajectory**: Frustrated then Angry then Threatening Churn **Timeline of Events** | Day | Event | Owner | Customer Sentiment | | --- | ----------------------------------------- | ---------- | ------------------ | | 1 | Customer reports data sync failures | Agent A | Frustrated | | 2 | Agent A requests logs, customer provides | Agent A | Cooperative | | 3 | Agent A escalates to Tier 2 | Agent B | Neutral | | 5 | Agent B says "known issue, no ETA" | Agent B | Angry | | 7 | Customer asks for workaround, no response | - | Very Angry | | 9 | Customer emails support manager | Agent C | Threatening | | 10 | Customer emails VP Support | VP Support | Churn Risk | | 12 | Current state: awaiting response | - | - | **Root Cause Analysis** **Technical Issue**: Data sync fails when records exceed 10K rows (undocumented limit) **Service Failures**: 1. 2-day gap with no response (Day 5-7) 2. "No ETA" response without workaround or escalation path 3. No proactive communication after identifying known issue 4. Customer had to escalate twice to get attention **What Customer Wants** 1. **Immediate**: Workaround to unblock their team 2. **Short-term**: Fix for the sync limit issue 3. **Long-term**: Confidence this won't happen again 4. **Unstated**: Acknowledgment that we dropped the ball **Draft Response** Subject: Personal Apology and Resolution Path for Sync Issues Dear John, I'm Sarah Chen, VP of Customer Success at \[Company]. I've personally reviewed your case and want to apologize for the experience you've had over the past 12 days. **You're right to be frustrated.** When you reported the sync failures on \[date], you deserved a faster, more transparent response. Instead, you received delays and a "no ETA" that left your team stuck. That's not the service level you signed up for, and I take full responsibility. **Here's what we're doing:** **Immediate (Today)**: Our engineering team has identified the root cause (a 10K row limit that wasn't documented). We've deployed a hotfix that increases the limit to 100K rows. Your syncs should now complete successfully. Can you confirm on your end? **This Week**: * We're assigning you a dedicated Technical Account Manager (TAM) who will be your single point of contact going forward * Your TAM will conduct a health check on your implementation to identify any other potential issues before they impact you **Service Credit**: To acknowledge the disruption to your operations, we're applying a 2-month service credit (\$30K) to your account. This will appear on your next invoice. **Ongoing**: * We're updating our documentation to clearly state system limits * We're implementing automated alerts when customers approach limits * I'm personally reviewing our escalation process to prevent similar gaps I'd welcome a call to discuss this directly. I'm available \[times] this week — please let me know what works for you. Again, I'm sorry we let you down. We'll earn back your trust. Sincerely, Sarah Chen VP Customer Success \[Direct phone] **De-escalation Framework Used** 1. **Acknowledge emotion**: "You're right to be frustrated" 2. **Take ownership**: "I take full responsibility" 3. **Explain root cause** (without excuses): Technical limit + service gaps 4. **Present solution** with timeline and accountability 5. **Rebuild trust**: Credit + TAM + process improvements 6. **Open door**: Direct contact, call offer *** ## Use Case 10 — Data: The "Dashboard Translator" **Mission**: Turn data visualizations into executive-ready narratives ### The Problem Data teams create beautiful dashboards, but executives want narratives, not charts. Translating "what the data shows" into "what it means and what to do" requires business context that analysts often lack. ### Plugin Architecture ```text theme={null} Plugin Name: dashboard-translator Components: - Slash Command: /data:translate-dashboard - Sub-agents: - chart-reader.md (extract data from visualizations) - trend-detector.md (identify patterns, anomalies) - insight-synthesizer.md (so-what, now-what) - benchmark-comparator.md (vs. targets, vs. industry) - Skills: - executive-communication/ (pyramid principle, BLUF) - industry-benchmarks/ (SaaS metrics, e-commerce, fintech) - statistical-significance/ (when changes matter) ``` ### Workflow Execution #### Step 1: User Invokes Command ```text theme={null} /data:translate-dashboard Q4-metrics-dashboard.png ``` (Or with raw data: `/data:translate-dashboard Q4-metrics.csv`) #### Step 2: Data Extraction Claude analyzes the dashboard image: * 4 line charts (ARR, MRR, NRR, Churn) * 2 bar charts (New vs. Expansion revenue, Sales by segment) * 3 KPI cards (ARR, Growth Rate, LTV:CAC) * 1 cohort table (retention by signup month) #### Step 3: Pattern Detection ```text theme={null} TRENDS IDENTIFIED: 1. ARR Growth: Accelerating - Q1: +8% QoQ - Q2: +10% QoQ - Q3: +11% QoQ - Q4: +12% QoQ Pattern: Consistent acceleration over 4 quarters 2. Segment Mix Shift: - Enterprise: 25% → 40% of new bookings - Mid-market: 50% → 45% - SMB: 25% → 15% Pattern: Successful upmarket motion 3. Cohort Retention: - 2023 cohorts: 85% 12-month retention - 2024 cohorts: 91% 12-month retention Pattern: Product improvements driving stickiness 4. ANOMALY: Q4 churn spike - October: 1.2% monthly churn (normal) - November: 2.8% monthly churn (elevated) - December: 1.4% monthly churn (normal) Investigation needed: November outlier ``` #### Step 4: Output Generation **Document: `Q4-Dashboard-Insights.docx`** **Executive Summary** **Bottom Line**: Q4 demonstrates strong momentum with accelerating growth and improving unit economics. One anomaly (November churn spike) requires investigation but doesn't change the overall positive trajectory. **Headline**: ARR growth accelerating; enterprise motion working; product stickiness improving. **Key Insights** **1. Revenue Growth Accelerating** **What the data shows**: ARR grew 12% QoQ in Q4, up from 8% in Q1. This is the fourth consecutive quarter of acceleration. **What it means**: Growth is compounding, not plateauing. The business is finding new levers (enterprise segment, expansion revenue) that supplement the core SMB motion. **What to do**: Continue investing in enterprise GTM. Consider raising growth targets for 2026 planning. Supporting data: | Quarter | QoQ Growth | YoY Growth | | ------- | ---------- | ---------- | | Q1 2025 | 8% | 28% | | Q2 2025 | 10% | 31% | | Q3 2025 | 11% | 33% | | Q4 2025 | 12% | 35% | **2. Enterprise Motion Working** **What the data shows**: Enterprise deals now represent 40% of new bookings, up from 25% a year ago. Average deal size increased from \$45K to \$78K. **What it means**: The upmarket push is succeeding. Enterprise customers have longer sales cycles but higher LTV and lower churn. **What to do**: Validate with cohort analysis that enterprise retention is indeed higher. If confirmed, consider reallocating more resources from SMB to enterprise. **3. Product Stickiness Improving** **What the data shows**: 2024 signup cohorts show 91% 12-month retention vs. 85% for 2023 cohorts. **What it means**: Product improvements (likely the new analytics dashboard and integrations shipped in 2024) are creating more value for customers. **What to do**: Identify which features correlate with retention. Double down on those in 2026 roadmap. **4. November Churn Anomaly** **What the data shows**: November monthly churn spiked to 2.8% vs. typical 1.2-1.5%. **What it means**: Something unusual happened in November. Could be: * Specific large customer(s) churned * Billing issue causing involuntary churn * Seasonal budget cuts * Competitive displacement **What to do**: 1. Pull list of November churned customers 2. Segment by reason (voluntary vs. involuntary) 3. If concentrated in specific accounts, conduct exit interviews 4. If billing-related, audit payment failure recovery process **Risks to Monitor** | Risk | Indicator | Current | Threshold | Status | | ------------------------ | -------------------------- | -------- | -------------- | ------ | | Enterprise concentration | Top 10 customer % of ARR | 35% | Under 40% | Watch | | Sales cycle lengthening | Days to close (enterprise) | 120 days | Under 100 days | Watch | | Pipeline coverage | Pipeline / Target | 2.6x | Over 3.0x | Watch | **Recommended Actions** 1. **Immediate**: Investigate November churn spike; report findings to leadership within 1 week 2. **Q1 Priority**: Build enterprise-specific cohort analysis to validate retention hypothesis 3. **2026 Planning**: Model scenarios with accelerated enterprise investment 4. **Dashboard Enhancement**: Add leading indicators (NPS trend, support ticket volume, feature adoption) *** ## Creating Custom Plugins — Full Walkthrough ### Example: The "Contract Cliff-Detector" **Mission**: Alert when contracts are approaching auto-renewal deadlines #### Step 1: Plan Your Plugin ```text theme={null} Use Case: Legal/Finance teams need to know which contracts are approaching auto-renewal deadlines so they can: - Renegotiate unfavorable terms - Cancel unwanted subscriptions - Budget for renewals Components Needed: - Slash command: /contracts:find-renewals - Sub-agent: Contract parser (extract renewal terms) - Skill: Auto-renewal clause patterns - Hook: Weekly scheduled scan (optional) - MCP: Google Calendar (create reminder events) ``` #### Step 2: Create Directory Structure ```bash theme={null} mkdir -p contract-cliff-detector/.claude-plugin mkdir -p contract-cliff-detector/commands mkdir -p contract-cliff-detector/agents mkdir -p contract-cliff-detector/skills/renewal-patterns cd contract-cliff-detector ``` #### Step 3: Create Manifest Create `.claude-plugin/plugin.json`: ```json theme={null} { "name": "contract-cliff-detector", "version": "1.0.0", "description": "Scan contracts for auto-renewal deadlines and alert before it's too late", "author": { "name": "Your Legal Team", "email": "legal@company.com" }, "keywords": ["legal", "contracts", "renewals", "compliance", "procurement"], "repository": "https://github.com/yourcompany/contract-cliff-detector" } ``` #### Step 4: Create Slash Command Create `commands/find-renewals.md`: ```markdown theme={null} --- description: Scan contracts for upcoming auto-renewal deadlines arguments: [days-ahead] --- # Contract Renewal Scanner Scan all contracts in the specified folder and identify those with auto-renewal clauses approaching their notice deadlines. ## Arguments - **Contract Folder**: Path to folder containing contracts (PDF, DOCX) - **Days Ahead** (optional): How far ahead to look (default: 90 days) User provided: $ARGUMENTS ## Process ### Step 1: Inventory Contracts List all PDF and DOCX files in the specified folder. Report total count to user. ### Step 2: Parallel Extraction For each contract, spawn a sub-agent (renewal-extractor) to find: - Contract parties (us and counterparty) - Contract type (SaaS, MSA, NDA, etc.) - Initial term end date - Auto-renewal clause (yes/no) - Renewal term length (e.g., 1 year, 3 years) - Notice period required (e.g., 30 days, 90 days) - Notice deadline (calculated: term end minus notice period) ### Step 3: Filter and Alert Filter contracts where: - Auto-renewal = Yes - Notice deadline <= today + [days-ahead] ### Step 4: Generate Output Create Excel workbook with: **Tab 1: Urgent Renewals** (notice deadline within 30 days) Columns: Contract, Counterparty, Annual Value, Notice Deadline, Days Remaining, Action Required Conditional formatting: Red if <14 days, Yellow if <30 days **Tab 2: Upcoming Renewals** (notice deadline within [days-ahead]) Same columns, sorted by deadline **Tab 3: All Contracts** Full inventory with renewal terms extracted **Tab 4: No Renewal Clause** Contracts without auto-renewal (for reference) Save as: contract-renewal-alerts.xlsx ### Step 5: Calendar Integration (if MCP available) For each urgent renewal, create calendar event: - Title: "CONTRACT RENEWAL DEADLINE: [Counterparty]" - Date: Notice deadline - Description: Contract details, action required - Reminder: 7 days before, 1 day before ``` #### Step 5: Create Sub-agent Create `agents/renewal-extractor.md`: ```yaml theme={null} --- name: renewal-extractor description: Extract renewal terms from a single contract model: sonnet-4.5 tools: [Read] permissions: file_access: read_only --- You are a contract analyst specializing in renewal terms. Your task is to extract auto-renewal information from a single contract. ## Instructions 1. Read the entire contract 2. Find the "Term" or "Duration" section 3. Look for auto-renewal language (see patterns below) 4. Extract specific values 5. Calculate notice deadline ## Common Auto-Renewal Patterns Look for phrases like: - "automatically renew for successive [period] terms" - "shall renew unless [party] provides written notice" - "evergreen clause" - "continuous renewal" - "unless terminated with [X] days notice" ## Return Format Return ONLY valid JSON: { "contract_file": "filename.pdf", "parties": { "us": "Our Company Name", "counterparty": "Vendor/Customer Name" }, "contract_type": "SaaS Agreement | MSA | NDA | SOW | License | Other", "effective_date": "2024-01-15", "initial_term": { "end_date": "2025-01-14", "length": "1 year" }, "auto_renewal": { "has_auto_renewal": true, "renewal_term": "1 year", "notice_period_days": 30, "notice_deadline": "2024-12-15", "exact_clause": "This Agreement shall automatically renew..." }, "financial": { "annual_value": 50000, "currency": "USD" }, "termination": { "can_terminate_for_convenience": true, "termination_notice_days": 30 } } If no auto-renewal clause exists: { "contract_file": "filename.pdf", "parties": {}, "auto_renewal": { "has_auto_renewal": false, "notes": "Contract expires on end date with no auto-renewal" } } ``` #### Step 6: Create Skill Create `skills/renewal-patterns/SKILL.md`: ```markdown theme={null} --- name: renewal-patterns description: Common auto-renewal clause patterns and terminology. Use when analyzing contracts for renewal terms. --- # Auto-Renewal Clause Patterns ## Standard Language Variations ### Pattern 1: Evergreen with Notice "This Agreement shall automatically renew for successive one (1) year terms unless either party provides written notice of non-renewal at least thirty (30) days prior to the end of the then-current term." Extract: - Renewal term: 1 year - Notice period: 30 days - Notice type: Written ### Pattern 2: Silent Renewal "Unless terminated in accordance with Section [X], this Agreement shall continue in effect indefinitely." Extract: - This is NOT auto-renewal, it's evergreen - Look for termination section for notice requirements ### Pattern 3: Price Escalation on Renewal "Upon renewal, fees shall increase by the greater of (i) 5% or (ii) the Consumer Price Index change for the prior 12 months." Extract: - Renewal terms: Standard - Flag: Price escalation clause (risk) ### Pattern 4: Multi-Year Commitment "Initial term of three (3) years, automatically renewing for additional three (3) year terms." Extract: - Renewal term: 3 years (long commitment!) - Flag: Multi-year auto-renewal (high risk) ## Notice Deadline Calculation Formula: Notice Deadline = Term End Date - Notice Period Days Example: - Term ends: December 31, 2025 - Notice period: 90 days - Notice deadline: October 2, 2025 ## Risk Flags Mark as HIGH RISK if: - Notice period > 60 days (easy to miss) - Renewal term > 1 year (long commitment) - Price escalation on renewal - No termination for convenience Mark as MEDIUM RISK if: - Notice period 30-60 days - Renewal term = 1 year - Notice must be "written" (vs. email OK) Mark as LOW RISK if: - Notice period < 30 days - Month-to-month after initial term - Email notice accepted ``` #### Step 7: Test Plugin ```bash theme={null} # Test locally before installing claude --plugin-dir ./contract-cliff-detector # In Claude session: /contract-cliff-detector:find-renewals /contracts/vendor-agreements/ 90 ``` #### Step 8: Install Plugin ```bash theme={null} # Install to user scope (personal use) claude plugin install ./contract-cliff-detector --scope user # Or install to project scope (share with team via git) claude plugin install ./contract-cliff-detector --scope project ``` #### Step 9: Distribute (Optional) **Option A: Share via Git** ```bash theme={null} git init git add . git commit -m "Initial contract-cliff-detector plugin" git remote add origin https://github.com/yourcompany/contract-cliff-detector git push -u origin main # Team members install with: claude plugin install https://github.com/yourcompany/contract-cliff-detector ``` **Option B: Submit to Plugin Marketplace** ```text theme={null} Coming soon: Anthropic plugin marketplace submission Follow https://docs.anthropic.com/plugins for updates ``` *** ## Real-World Plugin Examples ### Example 1: Sales "Prospect Research" **Workflow**: ```text theme={null} User: /sales:research-prospect "Acme Corp" Claude orchestrates: 1. MCP: Crunchbase → Funding: $15M Series A (2024) → Investors: Sequoia, a16z → Employee count: 120 → Growth: +50% headcount YoY 2. MCP: LinkedIn Sales Navigator → Decision makers: CTO, VP Engineering, VP Product → Recent hires: 5 engineers, 2 PMs in last quarter → Org chart: Engineering reports to CTO 3. Web scraping (via Browser MCP) → Product: B2B SaaS for inventory management → Pricing: Not public (request demo) → Tech stack: React, Python, AWS (from job postings) 4. MCP: Salesforce → Existing record? No → Past interactions? No → Similar customers? Yes - 3 inventory management companies ``` **Output: Prospect-Acme-Corp.docx** **Company Profile** | Attribute | Value | | ------------- | ------------------------------- | | Company | Acme Corp | | Industry | B2B SaaS - Inventory Management | | Stage | Series A (\$15M) | | Size | 120 employees | | Tech Stack | React, Python, AWS | | Growth Signal | +50% headcount YoY | **Key Contacts** | Name | Title | LinkedIn | Notes | | ---------- | -------------- | ------------- | --------------------------------- | | Jane Smith | CTO | /in/janesmith | Decision maker for dev tools | | John Doe | VP Engineering | /in/johndoe | Reports to CTO, 15 direct reports | | Sarah Lee | VP Product | /in/sarahlee | Owns roadmap, evaluation criteria | **Pain Points (Inferred from Job Postings)** 1. "Scaling challenges" mentioned in 3 engineering roles 2. "Data pipeline" issues — hiring 2 data engineers 3. "Manual processes" — looking for ops automation **Discovery Questions (MEDDIC Framework)** 1. Metrics: "How are you measuring deployment velocity today?" 2. Economic Buyer: "Who owns the budget for developer tools?" 3. Decision Criteria: "What's most important: speed, cost, or reliability?" 4. Decision Process: "Walk me through how you evaluated your last tool purchase" 5. Identify Pain: "You mentioned scaling challenges — tell me more" 6. Champion: "Who on your team would benefit most from solving this?" ### Example 2: Finance "Expense Categorization" **Workflow**: ```text theme={null} User: /finance:categorize-expenses /receipts/Q4-2025/ Claude processes: Folder contains: 127 receipt images (photos, PDFs, screenshots) Parallel processing (127 sub-agents): Each sub-agent: 1. OCR extracts text from receipt image 2. Identifies: Merchant, date, amount, line items 3. Categorizes per company expense policy 4. Flags policy violations ``` **Results aggregated — Output: Q4-Expense-Report.xlsx** **Tab 1: Categorized Expenses** | Date | Merchant | Category | Amount | Employee | Status | Notes | | ----- | -------------- | --------- | ------- | -------- | ------ | ------------------ | | 12/15 | Delta Airlines | Travel | \$450 | J. Smith | OK | SFO-NYC flight | | 12/15 | Marriott | Travel | \$289 | J. Smith | OK | NYC hotel | | 12/16 | Capital Grille | Meals | \$312 | J. Smith | Flag | Amount over \$100 | | 12/16 | Uber | Transport | \$45 | J. Smith | OK | Airport transfer | | 12/17 | Best Buy | Equipment | \$1,299 | M. Jones | Flag | Needs pre-approval | **Tab 2: Policy Violations** | Receipt | Violation | Policy | Action Required | | -------------------- | ------------------------------------- | ------------------------ | ------------------------------ | | Capital Grille \$312 | Meal exceeds \$100 limit | Meals Policy 3.2 | Manager approval + guest names | | Best Buy \$1,299 | Equipment over \$500 not pre-approved | Procurement Policy 2.1 | VP approval required | | Amazon \$89 | Missing receipt detail | Documentation Policy 1.4 | Itemized receipt needed | **Tab 3: Summary by Category** | Category | Count | Total | % of Total | Budget | Variance | | --------- | ------- | ------------ | ---------- | ------------ | -------- | | Travel | 34 | \$12,450 | 42% | \$15,000 | -17% | | Meals | 45 | \$4,230 | 14% | \$5,000 | -15% | | Software | 23 | \$8,900 | 30% | \$8,000 | +11% | | Equipment | 12 | \$3,200 | 11% | \$3,500 | -9% | | Other | 13 | \$890 | 3% | \$1,000 | -11% | | **Total** | **127** | **\$29,670** | **100%** | **\$32,500** | **-9%** | **Tab 4: Employee Summary** | Employee | Receipts | Total | Violations | Status | | -------- | -------- | ------- | ---------- | ------------- | | J. Smith | 45 | \$8,900 | 2 | Review needed | | M. Jones | 38 | \$7,200 | 1 | Review needed | | A. Lee | 28 | \$6,800 | 0 | Approved | | B. Chen | 16 | \$6,770 | 0 | Approved | ### Example 3: Product "Interview Synthesizer" **Workflow**: ```text theme={null} User: /product:synthesize-interviews /research/Q4-interviews/ Claude processes: Folder contains: 20 interview transcripts (60-90 min each) Parallel processing (20 sub-agents): Each sub-agent reads one transcript and extracts: - Jobs-to-be-done (what customer is trying to accomplish) - Pain points (frustrations, workarounds) - Feature requests (explicit asks) - Quotes (verbatim compelling statements) - Sentiment (overall satisfaction level) Synthesis phase: - Cluster pain points by theme - Rank by frequency x intensity - Identify patterns across segments ``` **Output: Customer-Interview-Synthesis-Q4.docx** **Executive Summary** 20 customer interviews conducted in Q4 2025. Customer segments: Enterprise (8), Mid-market (7), SMB (5). **Top 3 Themes:** 1. Reporting limitations (15/20 customers, 75%) 2. Mobile experience gaps (12/20 customers, 60%) 3. Integration reliability (10/20 customers, 50%) **Theme 1: Reporting Limitations** **Frequency**: 15/20 customers (75%) **Intensity**: High (8 mentioned as "major pain point") **Segment**: Strongest in Enterprise (7/8) Specific Requests: * Custom report builder (12 requests) * Scheduled report delivery via email (8 requests) * Export to Google Sheets, not just Excel (6 requests) * Role-based report access (4 requests) Recommendation: High priority for 2026 roadmap. Consider MVP with scheduled exports via email as a quick win, then V1 basic custom report builder, then V2 advanced builder with calculated fields and sharing. **Theme 2: Mobile Experience Gaps** **Frequency**: 12/20 customers (60%) **Intensity**: Medium-High (5 mentioned considering alternatives) **Segment**: Equal across all segments Specific Requests: * Mobile approval workflows (10 requests) * Android stability fixes (5 requests) * Offline mode (4 requests) * Mobile notifications that actually work (3 requests) Recommendation: Medium-high priority. Approval workflows are table stakes for enterprise. Android stability is reputation risk. **Theme 3: Integration Reliability** **Frequency**: 10/20 customers (50%) **Intensity**: High (7 described as "daily frustration") **Segment**: Strongest in Mid-market (6/7) Specific Requests: * Sync reliability improvements (8 requests) * Better error notifications (6 requests) * Self-service sync troubleshooting (4 requests) * Sync status dashboard (3 requests) Recommendation: Critical for retention. Integration issues directly correlate with churn risk. Prioritize reliability over new integrations. **Appendix: All Feature Requests (Ranked)** | Rank | Feature | Requests | Segment | Priority | | ---- | ------------------------- | -------- | ---------- | -------- | | 1 | Custom report builder | 12 | Enterprise | High | | 2 | Mobile approval workflows | 10 | All | High | | 3 | Integration reliability | 8 | Mid-market | Critical | | 4 | Scheduled report delivery | 8 | Enterprise | Medium | | 5 | Android stability | 5 | All | High | | 6 | Google Sheets export | 6 | Mid-market | Medium | | 7 | Error notifications | 6 | Mid-market | High | | 8 | Offline mobile mode | 4 | SMB | Low | | 9 | Role-based reports | 4 | Enterprise | Medium | | 10 | Self-serve sync tools | 4 | Mid-market | Medium | *** ## Key Takeaways ### Getting Started 1. **Start Small**: Install pre-built plugins before building custom ones. The official library covers most common use cases. 2. **Test Before Scaling**: Use `claude --plugin-dir ./your-plugin` to test locally before installing permanently. 3. **Leverage Community**: 9,000+ community plugins already exist. Search before building from scratch. 4. **Focus on High-ROI**: Target repetitive, time-consuming tasks that you do weekly or daily. Even 30-minute time savings compounds. 5. **Iterate**: Plugins can be customized after installation. Start with defaults, then refine based on your workflow. ### Architecture Decisions | Scenario | Recommended Approach | | -------------------------- | ------------------------------------------------ | | Process under 50 documents | Single Cowork session | | Process 50-500 documents | Cowork with parallel sub-agents | | Process 500+ documents | Batch API (Stage 1) + Cowork synthesis (Stage 2) | | Need external data | Add MCP server connections | | Need automation triggers | Add hooks for lifecycle events | | Need consistent behavior | Add skills for domain knowledge | ### Common Pitfalls | Pitfall | Solution | | ---------------------- | ---------------------------------------------------------------------------- | | Plugin doesn't load | Check manifest location: `.claude-plugin/plugin.json` | | Commands not appearing | Verify `commands/` directory is at plugin root, not inside `.claude-plugin/` | | Sub-agents failing | Ensure model specified in frontmatter is available (e.g., `sonnet-4.5`) | | MCP connection errors | Verify environment variables are set, credentials are valid | | Slow performance | Use parallel sub-agents, consider Batch API for high-volume | ### Next Steps 1. **Download Claude Desktop**: [https://claude.com/download](https://claude.com/download) 2. **Enable Cowork**: Click "Cowork" tab, grant folder access 3. **Install Your First Plugin**: Cowork tab, then Plugins, then Browse, then Install 4. **Try a Built-in Command**: Type `/` to see available commands 5. **Join the Community**: * GitHub: [https://github.com/anthropics/claude-plugins](https://github.com/anthropics/claude-plugins) * Discord: [https://discord.gg/anthropic](https://discord.gg/anthropic) * Docs: [https://docs.anthropic.com/plugins](https://docs.anthropic.com/plugins) *** ## Conclusion Claude Cowork plugins represent a fundamental shift in how knowledge workers interact with AI. Instead of generic conversations, you now have access to specialized, repeatable workflows that understand your domain. The key insight: Plugins aren't just about saving time (though they do — often 50-80% reduction in manual work). They're about **consistency** and **scalability**. A plugin that extracts RFP requirements does it the same way every time, catches the same edge cases, and produces the same structured output — whether it's your 1st RFP or your 100th. **What we covered**: * How plugins work (5 components: commands, sub-agents, MCP, hooks, skills) * How to install and customize pre-built plugins * How to build your own plugins from scratch * 10 real-world use cases with full implementation details * Best practices for architecture and distribution **What's next for you**: 1. Identify your "Monday morning dread" tasks — the repetitive work you procrastinate on or would benefit from delegating 2. Find or build a plugin that automates 80% of it 3. Reclaim those hours for work that actually requires human judgment The future of knowledge work isn't human vs. AI. It's human **with** AI, and plugins are the bridge. # The Claude and Nano Banana 2 AI Content Intelligence Stack Source: https://intelligence.madebydas.com/playbooks/claude-nano-banana-stack 3 tactical frameworks for retail and ecommerce — signal detection, insight-to-action dashboards, and AI campaign production at a fraction of the cost. 3 Tactical Frameworks for Retail & Ecommerce in 2026. *** > # Before You Dive In > > This playbook is dense by design. It is meant to be built with, not skimmed once. Bookmark it, set it up, come back every time you need to ship faster. > > **Need the whole system built and run for you?** > > That is what DAS does—signal detection, creative strategy, paid media execution, retention and lifecycle, and performance optimization as a *managed service*. We build the whole loop, running continuously, with a team that owns the outcomes, not the deliverables. > > If you need more than 20 distinct creative concepts a month and your current process can't deliver them, we should talk. → [**amlan@madebydas.com**](mailto:amlan@madebydas.com) *** # The Problem No One Is Honestly Talking About Most brands have a content operations problem dressed up as a creative problem. They commission content. It takes 3–6 weeks to produce. By the time it ships, the cultural moment it was built for has passed. The ad goes into rotation, performance is flat, and the team chalks it up to creative fatigue or algorithm changes. Then the cycle repeats. **The actual failure happens in three places:** **1. Signal blindness.** The brand doesn't know what content is gaining velocity in their category right now; not what's trending broadly, but what is *specifically* and *in-real-time* accelerating among their exact buyer. They're creating from intuition or last quarter's learnings. Both are stale by default. **2. Insight paralysis.** Data exists everywhere—ad dashboards, email metrics, social listening tools, review platforms—but it lives in silos. No one has a single view of what's resonating, what's rising, and/or what needs to be killed. Teams stay busy reacting to individual metrics instead of reading the full picture. **3. Production drag.** Even when the right insight surfaces, execution is slow. Briefing agencies, booking shoots, waiting on edits; the window closes before the asset ships. The insight that should have driven a winning campaign becomes a post-mortem bullet point. **The brands winning in 2026 have rebuilt the loop.** Signal informs creation in hours → insights are unified and actionable → production is fast enough to be responsive. And the whole system runs on a stack of AI tools that cost a fraction of what it replaced. This playbook gives you that content intelligence system in three layers. *** # Layer 1: Signal Detection **Know What's Gaining Velocity Before Your Competitors Do** *A framework for tracking not just what's trending, but what's accelerating within your specific category.* **The Core Concept** Trending is a lagging signal. By the time something is trending, the early movers have already captured the attention. What you actually need is **velocity** — content that is gaining momentum in your category before it peaks. The difference between a brand that looks prescient and one that always seems late is almost never budget or creative talent. It's their intelligence loop. One team finds the signal early. The other team finds it on TikTok's For You page when everyone else already has. This layer gives you a replicable system to be the first team, every time. **The 3-Signal Framework** Before you run any prompt or set up any workflow, understand what you're hunting for: **Signal Type 1 — Category Velocity** Content formats, topics, and aesthetics that are accelerating in your product category. Not trends broadly, but specifically what formats are getting shares and saves from buyers who look like your customer. **Signal Type 2 — Sentiment Drift** Shifts in how your audience talks about a problem your product solves. When the language changes—when "hydration" becomes "barrier repair" or "weight loss" becomes "body composition"—the brands that catch the drift early own the new vocabulary. Brands that don't catch it look out of touch within 18 months. **Signal Type 3 — Competitive White Space** What your direct and indirect competitors are not creating. This is often more valuable than knowing what's working for them. White space is where you get uncontested attention. **The Daily Signal Sweep (20-Minute Workflow)** **Tools:** Perplexity Pro, Claude (with web search on), TikTok Creative Center, Meta Ad Library **Step 1 — Category Velocity Scan (7 min)** Open Perplexity Pro and run this prompt once per day, rotating your category keyword: ```text theme={null} You are a cultural intelligence analyst. Search across TikTok, Instagram Reels, YouTube Shorts, and Reddit from the last 72 hours. Category: [YOUR PRODUCT CATEGORY — e.g., "women's activewear", "home fragrance", "functional beverages"] Find me: 1. The top 5 content formats gaining the most engagement velocity RIGHT NOW in this category (not just high view count — specifically content that is gaining speed, i.e., posted recently and accelerating fast) 2. The top 3 topics or narratives driving shares and saves (not likes — shares and saves signal intent) 3. Any emerging vocabulary or phrases being used by consumers that differ from brand language Format as: [Format/Topic] | [Platform] | [Why it's accelerating] | [Example hook or angle] ``` Save the output to a running document — a simple Notion table or Google Sheet with date, category, and findings works fine. You are building a pattern library over time. **Step 2 — Sentiment Drift Check (5 min)** Open Claude with web search enabled and run this monthly (set a calendar reminder): ```text theme={null} Search recent Reddit threads, TikTok comments, and review sites (Amazon, Trustpilot, Google Reviews) from the last 30 days for [YOUR PRODUCT CATEGORY]. I want to understand: 1. How consumers are currently describing the problem my product solves — exact language they use, not marketing language 2. Any new frustrations or desires that have emerged in the last 60-90 days that weren't common before 3. Words or phrases that are appearing frequently that brands in this space are NOT yet using in their marketing Return as a side-by-side: [Old Language] vs [Emerging Language], plus a list of 5 "new vocabulary" terms worth owning. ``` > *This prompt alone has surfaced content angles that outperformed existing creative by 2–4x for brands we work with. The words your customers use to describe their problem are your headline copy. Let them write it for you.* **Step 3 — Competitive White Space Audit (8 min)** Go to the Meta Ad Library. Search for your top 2–3 direct competitors. Filter by "Active" ads. Screenshot or note the formats they are running. Then run this in Claude: ```text theme={null} I've reviewed the current ad creative of the following competitors in [CATEGORY]: [COMPETITOR 1], [COMPETITOR 2], [COMPETITOR 3]. They are predominantly running: [describe what you saw — e.g., "UGC-style testimonials, lifestyle flat lays, before/after splits"] Given this landscape, identify: 1. What content formats are NOT being used that have proven traction in adjacent categories 2. What emotional angles or narratives are absent from this category's advertising 3. What would a challenger brand do to look completely different from this feed while still converting? Give me 5 specific content directions with rationale. ``` The output of this step is your creative brief for the week. Not a strategy doc. A brief. **What to Do With Layer 1 Output** Every signal sweep should end with one decision: **what are we making this week based on what we just found?** If you are running a lean operation, pick one signal and build one asset around it. If you have a content team, assign the top three signals with a 48-hour turnaround window. The point is that production decisions are now being driven by real-time intelligence, not a quarterly content calendar that was locked in before you knew what this week looked like. *** # Layer 2: The Insight-to-Action Dashboard **See Everything. Miss Nothing. Act Fast.** *A framework to build a unified, interactive intelligence hub that turns scattered data into a single actionable view.* **The Core Concept** Most brands are not short on data; they are short on synthesis. Your Meta dashboard tells you CTR. Your Klaviyo tells you open rates. Your Shopify tells you conversion rate. Your review platform tells you sentiment. Your social tools tell you reach. Your attribution tools tell you (usually) a bit of a different story. None of these talk to each other. And the person responsible for making content and campaign decisions is manually reconciling them — usually once a week, usually in a meeting, usually too late to act on anything meaningful. The Insight-to-Action Dashboard is not a new tool. It is a system you build with tools you likely already have. The goal is one place where the most important signals converge, with a weekly prompt that turns those signals into a specific action list. There are three ways to build it. Start with the one that matches where your team is right now. They escalate in automation and setup cost — but all three produce the same output. **What Goes Into the Dashboard** **Tier 1 — Performance Signals (pull weekly)** * Paid social: CTR by creative type, CPM trends, frequency, top and bottom performers by format * Email/SMS: Open rate by subject line type, click rate by content block, revenue per send * Organic: Top posts by saves + shares (not just reach), comment sentiment on top performers * Site: Landing page conversion rate, add-to-cart rate, top traffic sources **Tier 2 — Customer Intelligence (pull monthly)** * Review themes: What words appear most in 4–5 star reviews vs. 1–2 star reviews * Support ticket clusters: What are customers confused about or disappointed by * Post-purchase survey data: Where did they first hear about you, what almost stopped them from buying **Tier 3 — Competitive Intelligence (pull from Layer 1)** * Category velocity signals from your daily sweep * Competitor ad creative changes * Emerging vocabulary from sentiment drift check **Option A — Manual Paste (No Setup, Works Today)** Export CSVs from Meta, Klaviyo, and Shopify. Paste the data into Claude. Run the Weekly Synthesis Prompt below. Zero setup cost, about 15 minutes to pull and paste each week. The limitation: it's a push workflow is that you have to go get the data yourself. This is the right starting point if you want to validate the workflow before building any automation around it. Most teams who try it once automate it within two weeks. **Option B — Google Sheets as the Hub (Recommended Starting Point)** This is the core workflow. No engineering required, low maintenance, and it turns your data from something you go find into something that's ready and waiting for you. **Step 1 — Connect your data sources to a master Google Sheet** * **Meta Ads:** Use the native Meta Ads Google Sheets integration inside Meta Business Suite → Reports → Export to Sheets. Set it to auto-refresh weekly. If you're pulling from multiple paid channels (Google, TikTok, Pinterest), use **Supermetrics** (\~\$99/mo) or [**Funnel.io**](http://Funnel.io) — both connect all major ad platforms into a single Sheet with one refresh. * **Klaviyo:** Klaviyo's reporting dashboard has a native Google Sheets export. For automated weekly appending, use a Zapier trigger: "Campaign sent" → wait 48 hours → append stats to Sheet. This captures the 48-hour performance window before data stabilizes. * **Shopify:** Install the **Shopify Connector for Google Sheets** (free, in the Google Workspace Marketplace). Set it to pull revenue, conversion rate, and AOV on a weekly scheduled refresh. Each platform gets its own tab in the Sheet. You are not trying to make this look good. You are just getting the data in one place. **Step 2 — Build the Weekly Digest tab** Create one additional tab called "Weekly Digest." Use QUERY or IMPORTRANGE formulas to pull the key metrics from each source tab into a single clean table. The goal is a paste-ready summary — 10–15 rows, the metrics that matter, nothing else: | **Week** | **Channel** | **Metric** | **This Week** | **Last Week** | **Delta** | | -------- | ----------- | ---------- | ------------- | ------------- | --------- | | Wk 11 | Meta | CTR | 1.8% | 1.4% | +0.4% | | Wk 11 | Meta | CPM | \$18.40 | \$21.20 | -\$2.80 | | Wk 11 | Meta | Top Format | UGC Flash | — | — | | Wk 11 | Klaviyo | Open Rate | 42% | 38% | +4% | | Wk 11 | Klaviyo | Rev/Send | \$0.38 | \$0.29 | +\$0.09 | | Wk 11 | Shopify | CVR | 3.1% | 2.8% | +0.3% | | Wk 11 | Shopify | AOV | \$74 | \$68 | +\$6 | **Step 3 — Run the synthesis** Every Monday morning: open the Weekly Digest tab, copy the table, paste it into Claude with the Weekly Synthesis Prompt below. Total time once the Sheet is set up: under 5 minutes. **Option C — Fully Automated Push ([Make.com](http://Make.com) + Claude API)** This version removes you from the loop entirely. The data is pulled, synthesized, and pushed to your inbox or Slack before your Monday planning meeting — without you touching anything. **The architecture:** \[Meta / Klaviyo / Shopify] → \[Google Sheet auto-refresh] → \[[Make.com](http://Make.com): triggers Monday 7am EST] → \[Read Weekly Digest rows] → \[Claude API: Weekly Synthesis Prompt + data] → \[Push output → Slack / Notion / Gmail] **Build steps in [Make.com](http://Make.com):** 1. Create a new Scenario. Set the trigger to **Schedule → Every week → Monday 7:00am EST.** 2. **Module 1 — Google Sheets: Get Range.** Connect your Sheet and select the Weekly Digest tab range (e.g., A1:G15). This pulls the current week's metrics. 3. **Module 2 — Text Aggregator.** Format the rows into a clean string. Map each row as channel | metric | this\_week | last\_week | delta separated by line breaks. This becomes the data block in your prompt. 4. **Module 3 — HTTP: Make a Request.** POST to the Claude API. 5. **Module 4 — Output.** Choose one: Slack (Create Message), Gmail (Send Email), or Notion (Create Page). **Cost per automated run:** \$0.01–\$0.03 in Claude API credits. Effectively free. Make's free tier supports up to 1,000 operations per month — this scenario uses fewer than 10 per run. **Best for:** Brands running 3+ marketing channels simultaneously, teams where the weekly data review is a calendar bottleneck, or any operator who wants the synthesis waiting for them on Monday morning rather than building it themselves. **The Weekly Synthesis Prompt** This prompt works the same whether you're running Option A, B, or C. The only difference is how the data gets in front of it. ```text theme={null} I'm going to give you this week's performance data for [BRAND NAME]. Your job is not to summarize it — I can read numbers. Your job is to tell me what it MEANS and what I should DO about it. Here is the data: [PASTE METRICS — CTR, CPM, open rates, conversion rates, top/bottom performers] Tell me: 1. What is the single clearest signal in this data — what is working and why, in plain language 2. What is the single biggest leak — where are we losing and what's likely causing it 3. What should we make MORE of this week based on this data 4. What should we kill or pause 5. One specific test I should run in the next 7 days to answer the biggest open question in this data Format as an Action Brief, not a report. Each answer should be 2–3 sentences max. I need to be able to act on this in 30 minutes. ``` Run it every Monday. Share the output with your creative and media team before they plan the week. Your team stops guessing and starts building toward a measurable signal. **The Monthly Intelligence Report** Once per month, run a deeper synthesis using the full Tier 1 + Tier 2 + Tier 3 data set: ```text theme={null} You are a senior growth strategist reviewing [BRAND NAME]'s full content and performance intelligence for [MONTH]. Here is the full data package: [PASTE OR ATTACH] Produce a Monthly Intelligence Report in this format: WHAT WORKED (and why it worked, not just that it worked) WHAT DIDN'T (root cause, not surface metric) CATEGORY SIGNALS (what's happening in the market we should be responding to) CONTENT STRATEGY SHIFTS (what we do differently next month based on everything above) THE ONE BET (if we could only do one thing differently next month for maximum leverage, what is it and why) Be direct. Be specific. Cite the data. Do not hedge. ``` This report is your creative brief for the following month. It also becomes the artifact you bring into agency or team reviews to align on direction without a two-hour strategy meeting. *** # Layer 3: AI Campaign Production **Fast Enough to Be Relevant. Polished Enough to Convert.** *A framework for blending leading AI production technologies with cultural relevance and creative direction to unlock a new era of content at a fraction of the cost.* **The Core Concept** If Layer 1 tells you what to make and Layer 2 tells you what's working, Layer 3 is where you actually make it — fast. The production economics have changed permanently. A traditional commercial shoot costs \$2K–\$10K per day and yields roughly 20 usable assets. To give Meta and TikTok the 100 variants they need to find efficiency, you are spending \$25K+ in production before a single impression is bought. AI-native production in 2026 doesn't replace your brand. Your homepage hero, your flagship campaign, your packaging — those still require a human photographer with full creative control. But everything else — the 50 Meta variants, the email headers, the seasonal swaps, the TikTok-native scroll-stoppers, the carousel cards — all of that can now be produced in hours, not weeks. This layer covers the current production stack by asset type, with specific tool recommendations and prompts. **The 2026 AI Production Stack** **Static Image Production** Best tool: **NanoBanana 2** (Gemini 2.0 Flash Image, released Feb 26, 2026) Why: Native text rendering on packaging, 4K output, near-professional lighting control. The first image model that actually solves the legible label problem — which makes it viable for conversion ads, not just editorial. Use this for: Product hero shots, Meta static ads, email headers, Amazon PDP assets, seasonal variants, UGC-style creatives. **Short-Form Video Production** Best tools: **Kling 1.6** (character consistency, product integration), **Hailuo AI** (motion quality, cinematic output), **Veo 2 by Google** (photorealistic video, best for lifestyle and nature-adjacent content) Use this for: TikTok and Reels creatives, product demo videos, lifestyle b-roll, seasonal brand films **Prompt architecture for video (adapt to any of the above tools):** ```text theme={null} [SCENE DESCRIPTION]: [Describe the exact scene — setting, lighting, action, camera movement] [PRODUCT PLACEMENT]: [How and where the product appears — held, on surface, in motion, etc.] [MOOD/TONE]: [3 adjectives that describe the emotional register of the clip] [CAMERA DIRECTION]: [Static, slow push, handheld, aerial, macro, etc.] [STYLE REFERENCE]: [Describe a visual reference — "Glossier campaign aesthetic", "iPhone UGC, raw, no color grade", "luxury fragrance TV spot"] [DURATION]: [3–5 seconds for scroll-stopping cuts, 8–15 seconds for narrative] [ASPECT RATIO]: [9:16 for TikTok/Reels, 4:5 for Meta feed, 16:9 for YouTube pre-roll] [NEGATIVE PROMPTS]: [What you don't want — "no text overlays, no talking heads, no stock footage feel"] ``` **Voice and Audio Layer** Best tools: **ElevenLabs** (voiceover, character voice cloning), **Suno** (brand-appropriate background music), **Adobe Podcast** (audio cleanup for raw UGC) Use this for: Adding voiceover to AI-generated video clips, background music for Reels, cleaning up real UGC audio before repurposing **Video Editing and Assembly** Best tools: **CapCut** (fast assembly, TikTok-native features), **Descript** (transcript-based editing, repurposing long-form), **Runway** (compositing, background removal, style transfer) Use this for: Assembling raw AI clips into finished ads, repurposing long-form video into short-form cuts, adding captions and text overlays **The Campaign Sprint Workflow** This is the full production workflow from signal to shipped asset. Designed to run in under 4 hours for a standard 10-asset campaign batch. **Phase 1 — Brief (20 min)** Use Layer 1 output (your daily signal sweep) to identify the creative angle. Answer three questions before you open any production tool: * Who is seeing this? (Which segment of your buyer — first-time visitor, lapsed customer, loyal buyer) * What do they need to believe? (The one conviction this creative needs to create) * What do we want them to do? (The single action — click, swipe up, add to cart) Write these three answers down. This is your brief. Everything you produce in Phase 2 should answer all three. **Phase 2 — Static Asset Production (60 min)** 1. Run your Brand DNA prompt in Claude (from LM1, Part 2) if you haven't already. This is a one-time setup. 2. Select 5 prompt types from the NanoBanana 2 playbook that match your brief 3. Prepend your Brand DNA IMAGE GENERATION PROMPT MODIFIER to each prompt 4. Generate 5 outputs per prompt type (the Rule of 5 — generate 5, keep 1) 5. Run the Quality Gate: text accuracy, product geometry, label legibility, 3-second scroll test Output: 5 production-ready static assets **Phase 3 — Video Asset Production (90 min)** 1. Select 2–3 of your static heroes from Phase 2 as reference frames 2. Write video prompts using the architecture above for the same scenes, extended into motion 3. Generate in Kling or Hailuo (run both if you have access — they have different strengths) 4. Select the best motion clip per scene 5. Add audio layer: ElevenLabs voiceover or Suno background track 6. Assemble in CapCut with captions Output: 2–3 short-form video assets, 9:16 for TikTok/Reels **Phase 4 — Quality Check and Platform Prep (30 min)** * Resize all assets to platform specs (9:16, 4:5, 1:1, 16:9) * Verify all text overlays are legible at mobile size * Check compliance: no exaggerated claims, no fake UI elements, no policy violations * Name files with a convention that tracks the variant (e.g., \[Brand]*\[Format]*\[Angle]*\[Platform]*\[Date]) **Phase 5 — Launch and Loop Back (20 min)** * Upload to your media buyer or ad manager * Tag creative in your tracking system by format, angle, and audience segment * Set a Layer 2 reminder for 72 hours post-launch to pull initial CTR and thumb-stop data * Feed that data back into next week's Layer 1 signal sweep Total time: Under 4 hours from brief to uploaded assets. **The Tier System (What AI Replaces vs. What It Doesn't)** Not everything should be AI-generated. Being clear about this is what separates brands that use this system well from brands that use it to erode their visual identity. | **Tier** | **Asset Type** | **Production Method** | | -------- | ------------------------------------------------------ | ----------------------------------- | | 1 | Homepage hero, campaign flagship, packaging | Human photographer, full production | | 2 | Meta/TikTok ad variants, email headers, seasonal swaps | AI production (this playbook) | | 3 | Thumbnails, carousel cards, story assets, PDP badges | AI production, faster workflow | Tier 1 assets are what your brand is. Tier 2 and 3 are what your brand does at scale. Both matter. The mistake most brands make is either spending Tier 1 resources on Tier 2 and 3 assets, or using Tier 2 tools on assets that needed Tier 1 treatment. *** # The 30-Day Stack Sprint Here is how to implement all three layers in 30 days without overhauling your entire operation. **Week 1 — Signal Infrastructure** * Day 1–2: Set up your Daily Signal Sweep workflow. Run it every morning for five days. Do not act on it yet — just observe what you find. * Day 3–5: Run the Competitive White Space audit. Identify the three content directions your competitors are ignoring. * Day 5–7: Pick one white space direction. Brief one asset against it (static, video, or email). Ship it by end of week. **Week 2 — Intelligence Dashboard** * Day 8–9: Pull your last 30 days of Tier 1 performance data into one document. * Day 10: Run the Weekly Synthesis Prompt. Read the output with your media buyer or content lead. * Day 11–14: Use the Action Brief output to make two decisions: what to scale, what to cut. Execute one of each. **Week 3 — Production Stack Setup** * Day 15: Run the Brand DNA prompt in Claude. Save the output as your brand's permanent prompt modifier. * Day 16–17: Generate your first batch of 10 static assets using NanoBanana 2 and the prompt templates. Run the full Quality Gate. * Day 18–19: Produce your first AI video asset. Assemble in CapCut. Add audio. * Day 20–21: Launch the full batch. Tag all creative in your tracking system. **Week 4 — Loop and Refine** * Day 22–24: Pull 72-hour data on Week 3 creative. Run the Weekly Synthesis Prompt with this data. * Day 25–27: Identify your top-performing asset. Generate 10 variants of it using seed locking (keep the composition, change one variable at a time). * Day 28–30: Document what worked. Write a one-page "Stack Brief" capturing your brand's signal patterns, top-performing formats, and production workflow. This becomes the operating manual for everyone who touches content at your brand. *** # Closing This stack exists because the brands that figure out how to move at the speed of culture — not the speed of production — are the ones that own the next decade of retail attention. The three layers are not sequential. They are a loop. Signal informs production → Production generates data → Data informs the next signal sweep. The brands running this loop daily are not just faster, they are compounding a creative intelligence advantage that is very hard to replicate once it's established. If you want this built for your brand — custom signal workflows, dashboard setup, or a full production sprint — reach out. [**amlan@madebydas.com**](mailto:amlan@madebydas.com) [**newbiz@madebydas.com**](mailto:newbiz@madebydas.com) # The CMO's AI Guide to Customer Intelligence: 25 Principles Source: https://intelligence.madebydas.com/playbooks/cmo-ai-guide 25 principles for CMOs at retail and ecommerce brands who want compounding results from AI — not just faster campaigns. Covers why 95% of CMOs using AI see no business outcome gains, how customer intelligence provides the foundation AI needs, and the specific operational frameworks that separate the 5% from everyone else. Published by DAS — The Audience Development Agency. # The CMO's AI Guide to Customer Intelligence ## 25 Principles for Marketing Leaders Who Want Compounding Results The core finding: 75% of marketers have adopted AI. 84% are still running generic campaigns. The problem isn't adoption — it's application. AI pointed at tactics without a foundation of customer understanding produces more slop, faster. This guide is for the CMOs who already know that. **Why this exists:** Gartner found that only 5% of CMOs using generative AI are seeing significant gains on business outcomes. The other 95% are using AI to move campaigns faster. The ones in the 5% are using it to understand their customers first — and then executing. The sequence is everything. | Stat | Source | | -------------------------------------------------------------------------- | --------- | | Companies that excel at customer understanding generate 40% more revenue | McKinsey | | Customer-obsessed companies grow revenue 2.5× faster than peers | Forrester | | Only 5% of CMOs using gen AI are seeing significant business outcome gains | Gartner | Customer intelligence first. Campaigns second. Everything else is noise. *** ## Part I — The Mindset Shift: Before You Touch a Tool *Five principles that separate compounding from resetting* ### 01 — AI won't fix a broken customer strategy If you don't know which customers are actually profitable, AI will help you reach more unprofitable ones faster. Every tool in your stack — Klaviyo, Meta, Claude — amplifies your existing strategy. If that strategy starts with acquisition before understanding, AI accelerates the wrong direction. **→ Do this:** Before your next campaign brief, answer one question: What is the 12-month contribution margin of our top 20% of customers versus our bottom 20%? If you can't answer that, that's the gap. Not your tool stack. ### 02 — Your customer file is the real competitive moat — not your tool stack Every brand on Shopify Plus has access to Claude, ChatGPT, and Klaviyo. What they don't have is your customer data. The brands winning with AI are feeding proprietary behavioral intelligence into general-purpose tools. The tool is commoditized. The intelligence is not. **→ Do this:** Ask honestly — what do we know about our best customers that competitors don't? If the answer is "nothing distinctive," your competitive advantage is a media budget. That won't compound. ### 03 — The compounding advantage is real — and it's widening Brands that built customer intelligence infrastructure in 2023–2024 aren't just ahead. They're accelerating. Each week of analysis makes next week's decisions smarter. Each campaign informs the next brief. Laggards cannot close this gap with a software purchase. They can only start building now. **→ Do this:** Start a rolling "Customer Intelligence Brief" — a Notion page your team appends to every Monday. In six months, it becomes the most valuable internal document you own. ### 04 — Stop measuring success with metrics that can't prove business impact Open rates tell you who opens. Platform ROAS tells you what the platform wants you to believe. The metrics your CFO cares about: * Contribution margin by channel * 90-day repeat purchase rate * Marketing Efficiency Ratio (total revenue ÷ total marketing spend) * Blended CAC with payback period * Champion segment size and revenue share **→ Do this:** Build a one-page weekly dashboard with those five metrics. Review it Monday morning before you open Klaviyo or your ads manager. ### 05 — AI makes mistakes. Your job is judgment — not verification theater Don't avoid AI because it hallucinates. Don't trust it blindly because it sounds authoritative. The productive stance: use AI to get to 80% faster, then apply domain expertise to the final 20%. **→ Do this:** Define two categories for your team: AI outputs that can go straight to execution, and AI outputs that require human review before acting. Write it down. That discipline separates fast from reckless. *** ## Part II — What to Understand First: Your Customer File *The intelligence that should precede every campaign brief* ### 06 — Map your RFM distribution before your next campaign brief Recency, Frequency, Monetary: these three dimensions reveal everything about who's actually driving your revenue. Klaviyo's built-in RFM scoring creates six groups — Champions, Loyal, Recent, Needs Attention, At Risk, Inactive — each requiring fundamentally different treatment. Most brands send all six groups the same message. That's not a personalization problem. It's an intelligence problem. **→ Do this:** Log into Klaviyo → Analytics → Customer Lifetime Value → RFM Groups. Note the size of each group and the % of revenue from Champions. Bring this number to your next campaign meeting. ### 07 — Your top 8% of customers are driving roughly half your revenue The Pareto distribution in ecommerce is steeper than most CMOs realize. | Segment | Revenue share | | --------------------------------------------- | ---------------- | | Top \~8% of customers (Champions) | \~50% of revenue | | Existing customers vs. new — spend difference | +67% more | | Loyal customers months 31–36 vs. months 0–6 | +67% more | Probability of selling to an existing customer: 60–70%. Probability of selling to a new prospect: 5–20%. **→ Do this:** Export your top 500 customers by LTV. What do they have in common — category, acquisition source, first product purchased, geography — that your current acquisition targeting doesn't reflect? ### 08 — The second purchase is the most important transaction in your business 70–80% of new DTC customers never return after their first purchase. The economics of improving second-purchase conversion dwarf any incremental improvement you'll achieve in new customer acquisition rate. A 5% increase in retention can boost profits by 25–95%. (Bain & Company) No acquisition optimization has that leverage ratio. **→ Do this:** Find your 90-day repeat purchase rate for first-time buyers. If you don't have that number, find it today. Then ask: does our post-purchase sequence actually drive second purchase — or does it just celebrate the first? ### 09 — Your customers are writing your best marketing copy in their reviews Export your last 500 reviews. Feed them into Claude. Ask it to identify the exact language customers use to describe transformation, pain points, objections, and purchase triggers. You'll find creative angles your team has never considered — because they were written by customers, not marketers. **→ Do this:** Export all reviews to a text file. Upload to Claude with this prompt: "Identify the top 15 emotional triggers and specific phrases customers use to describe this product. Group by sentiment and purchase trigger. Flag any phrases that appear verbatim 3+ times." Require at least 3 real customer phrases to appear verbatim in your next creative brief. ### 10 — Understand where your contribution margin actually lives before you optimize anything Platform ROAS doesn't capture subscription LTV, Amazon halo effects, or SKU-level margin variation. A 4.0 ROAS on a 15% margin product in a high-return category can be losing you money. Before you optimize any channel, map contribution margin by SKU and acquisition source. **→ Do this:** Ask your team: which of our top 5 acquisition channels produces the best 12-month contribution margin per customer — not ROAS, not CAC, but actual margin retained per cohort? If no one can answer, that's what you build next. *** ## Part III — How to Use AI for Customer Intelligence *Specific tools, workflows, and prompting approaches* ### 11 — Use Claude for deep analysis. ChatGPT for speed. Perplexity for research. | Tool | Best for | | ------------------------ | ---------------------------------------------------------------------------------------- | | Claude (200K+ context) | Processing large customer data files, review mining, strategic analysis, creative briefs | | ChatGPT Code Interpreter | Ad hoc data analysis, visualization, quick calculations | | Perplexity | Competitive research with citations, category intelligence, sourced market data | Optimal workflow: Perplexity for research → Claude for analysis → ChatGPT for execution and visualization. ### 12 — Build a Master Context Document — and feed it to every AI session The gap between generic AI output and genuinely useful intelligence is context. Maintain five living documents and upload them to every strategic AI session: 1. Brand DNA — voice, values, ICP definition, key messaging pillars 2. Customer Intelligence Brief — rolling weekly insights, RFM distributions, behavioral patterns 3. Competitive Landscape — competitor profiles, positioning, known strategies 4. Product Knowledge Base — full catalog, margins, seasonal patterns 5. Historical Performance Data — past campaign summaries, seasonal benchmarks, what worked and what didn't Claude's Projects feature stores these persistently across sessions. ### 13 — Most teams use less than 30% of the Klaviyo AI features already in their subscription | Feature | What it does | | ------------------------- | ---------------------------------------------------------- | | Predictive Analytics | Forecasts CLV, next order date, churn risk per customer | | Segments AI | Build segments from plain-English descriptions | | Personalized Campaigns AI | Sends each contact the version they'll respond to | | Smart Send Time | Optimizes delivery timing per individual recipient | | Churn Risk Scoring | Classifies each customer as Low / Medium / High churn risk | **→ Do this:** Turn on Predictive Analytics in Klaviyo. Identify your top 200 customers by predicted CLV. Create a "Predicted Champions" segment. This is the highest-value seed list you have for both retention and paid media lookalikes. ### 14 — Intervene at Medium churn risk — not High Klaviyo classifies customers as Low, Medium, or High churn risk. Critical finding: 88–97% of "Medium" risk customers actually churn. By the time a customer hits "High," most intervention windows have already closed. Treat Medium as your signal to act. **→ Do this:** Filter your Klaviyo list for Medium churn risk customers with CLV above your average. Multiply the count by average LTV. That number — the revenue you'd lose if you did nothing — is your business case for building the winback flow today. ### 15 — Run competitor review analysis like a product strategist, not a marketer Export competitor reviews from Amazon or their DTC site. Feed into Claude with this prompt: "Analyze these 500 reviews of \[Competitor]. Identify: top praised features, top complaints, unmet customer needs, and any mentions of alternative brands. What are customers clearly not getting from this product that they want?" This is market research that used to cost \$20,000 and take six weeks. ### 16 — Build a rolling weekly Customer Intelligence Brief **Weekly cadence:** * Monday (60 min): Export Klaviyo segment performance + Shopify order data. Upload to Claude. Run week-over-week trend analysis. Identify 2–3 action items. Append to the rolling doc. * Monthly (half-day): Full cohort analysis — retention curves, LTV by acquisition channel, RFM distribution shifts. * Quarterly (full day): Customer file deep dive, board-ready metrics, competitive landscape update. ### 17 — Use AI to build your creative brief from customer language — not brand language Most creative briefs are written in the brand's vocabulary. The best ones are written in the customer's vocabulary — the exact phrases, metaphors, and emotional triggers that appear in reviews, support tickets, and post-purchase surveys. AI makes the extraction fast. The insight is that you should be doing it at all. ### 18 — Automate competitive intelligence in 80 minutes a week | Day | Time | Task | | --------- | ------ | -------------------------------------------------------------------------------------------- | | Monday | 30 min | Perplexity queries on competitor/category moves. Meta Ad Library scan. Google Alerts review. | | Wednesday | 20 min | Competitor keyword and traffic shifts. Quick social scan. | | Friday | 30 min | Compile findings in Claude. Update competitive section of Intelligence Brief. | *** ## Part IV — High-Impact Use Cases *Specific applications that directly move margin and LTV* ### 19 — Use your top-LTV customers as the seed list for paid media lookalikes Using your top 10–25% of customers by LTV as lookalike seeds produces up to 26% lower CPA versus interest targeting. Optimal seed size: \~2,000 customers. Refresh every 7–14 days. For Meta value-based lookalikes: include a lifetime value column — Meta weights toward your highest-value customer profile. ### 20 — Stop celebrating ROAS. Start measuring incrementality. Analysis of 225 incrementality studies across DTC brands in the \$15M–\$100M range found no consistent relationship between platform-reported ROAS and true incremental lift. In 70%+ of agency audits, branded search contaminates "prospecting" campaigns — with 60–70% of conversions coming from people who would have searched the brand name regardless. **The math that should terrify you:** A 3:1 ROAS can represent a net loss. \$900 revenue on \$300 ad spend — minus \$1,005 in COGS, shipping, payment processing, and platform fees = –\$105 net. ### 21 — Treat your RFM segments as different customers — because they are | Segment | What they need | What they don't need | | --------- | ----------------------------------------------------- | -------------------------------- | | Champions | Early access. Exclusivity. Recognition. | A discount. They'll buy anyway. | | Loyal | Cross-sell. Loyalty upgrade. Small reward for action. | The same email as everyone else. | | At Risk | A reason to return. Social proof. Value signal. | A generic "We miss you" | | Inactive | One final attempt — then suppression. | 12 more emails they'll ignore. | Segmented emails return 3× the revenue per recipient of unsegmented lists (\$0.19 vs. \$0.06 RPR). ### 22 — Build churn interventions by customer value × risk — not flat winback flows | | High Churn Risk | Low Churn Risk | | -------------- | ------------------------------------------------------------------------------------------ | --------------------------------------------------------------------- | | **High Value** | Immediate personal outreach. Exclusive offer. Dedicated support contact. | VIP loyalty upgrade. Referral incentive. Early access to new product. | | **Low Value** | Automated winback. Progressive discount. Revenue threshold determines whether to continue. | Standard nurture. No special intervention needed. | ### 23 — Use cohort intelligence for seasonal planning — not just execution speed The seasonal question isn't "what promotions should we run?" It's "which of our customer cohorts will respond to what, based on their behavioral history?" Champions respond to early access and exclusivity. At-Risk customers respond to value and social proof. New customers in their first 90 days respond to education and trust signals. *** ## Part V — What to Stop Doing Immediately ### 24 — Stop over-discounting customers who would have bought anyway 40%+ of customers in winback flows who receive large discounts would have returned without them. Discounting Champions trains them to wait for offers. RFM data tells you which customers need a financial reason to buy and which ones just need a reason to feel valued. Discounts are a tool. Applied without intelligence, they're margin destruction dressed as retention. ### 25 — Stop trusting platform ROAS as your primary budget allocation signal Post-iOS 14.5, with an ATT opt-in rate of approximately 13.85%, 86% of iOS users opt out of tracking. Platform-reported ROAS is increasingly a self-reported metric, designed by platforms to justify their own existence. The CMO who allocates budget based on platform dashboards is making decisions optimized for the platform — not for the business. *** **The pattern across all 25 principles:** Every principle in this guide points to the same root issue — campaigns executing faster than the intelligence that should be driving them. The brands getting compounding results from AI aren't using better tools. They're asking better questions — of their data, their customers, and their assumptions. AI makes answering those questions faster. It doesn't make asking them for you. *A small group of your customers is driving most of your value. You're probably treating them like everyone else.* Click **Open in Claude** above to work through any of these 25 principles with your own customer data. Claude will have this full guide as context. # The CPG Brand Intelligence Playbook Source: https://intelligence.madebydas.com/playbooks/cpg-brand-intelligence How Gemini 3.1 Pro exposed the 6 dimensions separating the top 5% of wellness brands — product page structure, content velocity, email capture, social engagement, paid media efficiency, and brand narrative consistency. ## How Gemini 3.1 Pro Exposed the 6 Dimensions That Separate the Top 5% of Wellness Brands From Everyone Else *** ## Part 1: Why We Stopped Trusting Manual Brand Audits For the last decade, brand auditing was a problem of human bandwidth; you simply cannot manually process the data density required to benchmark a category. I recall sitting with a client doing \$64M GMV in the functional beverage space; we were trying to map the attribute density of their top 10 competitors. The cognitive load of evaluating 200+ brands across 6 distinct dimensions is impossible for a human analyst to maintain without degradation. Manual audits are low-resolution by definition. You pull up a Product Detail Page (PDP); check the meta descriptions; review the creative assets; scan the reviews. It is functional work, but you miss the subtle patterns. You miss the attribute density ratios in the bullets; you miss the claim stacking hierarchy; you miss the narrative fragmentation between the ad layer and the landing page. We shifted our methodology entirely with the release of Gemini 3.1 Pro on February 19, 2026. The model scored 77.1% on ARC-AGI-2 -- roughly 2x its predecessor -- but the reasoning capability wasn't the primary utility for us. The utility was the 1M token context window. We could finally hold an entire brand's digital footprint -- PDPs, social comments, ad libraries, email flows -- in a single, active session. This allows for a three-tier thinking system that traces connections a human operator would fatigue before finding. This isn't about replacing human judgment; that will always be crucial. This is more about expanding the resolution of the analysis. We ran this process on 200 CPG wellness brands ranging from \$35M to \$500M GMV. We scraped public digital footprints including Amazon PDPs, direct-to-consumer product pages, social profiles, email signup flows, and ad libraries. We fed this unstructured data into Gemini 3.1 Pro to score each brand against 6 specific dimensions. The results exposed a stark divide. The top 5% of brands weren't just "better" at marketing; they are fundamentally structured differently at the data layer. This guide delivers those 6 patterns. I have included the exact methodology, the benchmarks we observed, and the copy-paste prompts we used to extract the data. *** ## Part 2: Dimension 1 -- Product Page Structure Most brands treat their PDP as a digital brochure; the top 5% treat it as a database for retrieval. The data supports this shift; 78% of grocery searches on Amazon are now unbranded. Shoppers do not search for the brand name; they search by attributes. NIQ data indicates that products with verified claims grew +170 bps faster over a 5-year period compared to the category average. More interestingly, products utilizing multiple claim types grew approximately 2x faster than single-claim products. We observed a specific behavior in the top 5% of brands: **Attribute Density.** They stack searchable attributes -- "Organic," "Keto," "Non-GMO," "Wild-Caught" -- directly into the title, bullets, and description. They do not rely on image text; they ensure value propositions are machine-readable. This is critical for **Agent-Readiness**. As AI shopping agents become the primary interface for replenishment, the agent must be able to select the product based on text parameters alone. If your value prop is locked in a JPEG, the agent cannot read it; you are invisible to the machine. We also tracked the impact of A+ Content structure. A+ sections utilizing educational content saw 70% higher retention rates; use-case focused modules boosted conversions by up to 20%. Consider the difference in these two structures we analyzed: **Weak PDP:** "Premium Collagen Powder -- Feel Your Best." This relies on subjective sentiment; it fails the specificity test. **Strong PDP:** "Hydrolyzed Marine Collagen Peptides -- 10,000mg Type I & III -- Clinically Studied for Joint & Skin Support -- Wild-Caught, Non-GMO, Keto-Friendly." This is a high-resolution signal; it captures traffic for specific attribute searches. *** ## Part 3: Dimension 2 -- Content Velocity By late 2026, we project 65% of grocery sales will be digitally influenced. Content is the pre-purchase research layer that is the prerequisite. 'Brand awareness' content will need to be re-evaluated. When we audited the top performing brands, we defined **Content Velocity** differently than most agencies. Velocity is not just posting frequency. Velocity = Cadence + Type Mix + Channel Distribution + Production Speed. The top 5% of brands maintain a cadence of 4-5 posts per week per primary channel. More importantly, they adhere to a strict ratio: 70% educational / 30% promotional. They understand that the platform algorithms have shifted. LinkedIn's 2026 algorithm, for instance, heavily rewards dwell time, saves, and topical authority; it penalizes "broetry" and low-value engagement bait. For CPG, IG is the trade channel; TikTok is the discovery channel; Instagram is the community channel; Email is the retention channel. The most critical finding in this dimension was the compounding effect. Brands that maintained this velocity for 90+ days saw a non-linear increase in algorithmic authority. "Burst-and-fade" brands -- those that post heavily for a month then drop off -- never hit the compounding curve. *** ## Part 4: Dimension 3 -- Email Capture Mechanics DTC channels are predicted to generate 50% of overall CPG revenue by the end of 2026. With privacy regulations limiting third-party identifiers, the brands that win are the ones that own their audience data. The top 5% of brands in our analysis treat email as infrastructure, not a marketing channel. The differentiator is the **Value Exchange**. We saw one brand lift their capture rate by 22% simply by summarizing the value proposition in 15 words rather than a generic "Sign up for updates." They moved from "Join our newsletter" to "Get the 5-day Anti-Inflammatory Meal Plan." Segmentation from day one. The best brands use quiz funnels, preference selectors, or behavioral triggers immediately upon capture. They do not dump everyone into a general "Welcome" flow. (Please, please don't do this.) For wellness brands, the most effective lead magnets were utility-based: ingredient guides, transparency reports, or personalized routine builders. Consumers are willing to trade this data for relevance; 72% stated they would pay more for wellness products if they felt the brand understood their specific health goals. Post-capture, the top brands deploy a nurture sequence of 7-14 days focused purely on education before attempting a hard sell. In a cookie-deprecation world, the owned audience is the only leverage you have left. *** ## Part 5: Dimension 4 -- Social Engagement Quality Vanity metrics do not predict revenue for CPG wellness brands. We stopped tracking follower counts years ago; they are largely irrelevant to GMV. What matters is the **Conversation-to-Pipeline** pattern. Our analysis of the top 5% showed a distinct focus on "Conversation Triggers." These are posts specifically designed to generate questions and Direct Messages (DMs), rather than likes. TikTok Creative Center data corroborates this; UGC-style content outperforms polished brand creative by 2-3x in conversion because it invites dialogue. The top brands operate as communities, not broadcasters. The operational metric that defined the winners was **Response Velocity**. If a customer asks a question about ingredients or usage, 48+ hours unanswered equals a lost conversation. The top 5% of brands maintain a response protocol of under 2 hours during business hours. They also leverage Spark Ads on TikTok effectively. By boosting organic content that already has high engagement, they drive CPAs down by 30-50% compared to dark posts. *** ## Part 6: Dimension 5 -- Paid Media Efficiency The era of micro-targeting is effectively over; creative quality has replaced audience segmentation as the primary lever for performance. Meta's Q3 2025 Marketing Summit data confirms this shift: brands maintaining or improving ROAS did so through higher creative testing velocity, not granular audience exclusions. The top 5% of brands test 10-20 new ad variations per week; the median tests 2-3. More variants equal faster signal. For DTC e-commerce, the blended ROAS benchmarks are rigid. A healthy P\&L requires a blended ROAS of 3-4x. Within that mix, prospecting generally sits at 2-3x, while retargeting -- often overstated by platform attribution -- shows 6-10x. Platform mechanics favor automation over manual control. Meta's Advantage+ Shopping campaigns (ASC), utilizing broad targeting, consistently outperform manual setups by 15-25% in ROAS. → **TikTok:** Polished brand ads fail here. Lo-fi, UGC-style creative outperforms high-production assets by 2-3x in conversion. CPMs at \$4-\$7; CPCs at \$0.35-\$1.00. → **Category nuance:** Beverage brands iterate fastest; wellness requires longer compliance lead times; beauty/skincare relies on evergreen influencer whitelisting. Attribution remains the primary blind spot. Last-click attribution overvalues retargeting. We advise moving toward incrementality testing or Marketing Mix Modeling (MMM) to view true lift. *** ## Part 7: Dimension 6 -- Brand Narrative Consistency Consumers do not read strategy decks; they experience fragments of your brand across disconnected surfaces. The top 5% of brands maintain a single narrative thread across owned, earned, and paid channels. The median brand runs 3-4 different stories simultaneously. The PDP claims "premium science-backed clinicals"; TikTok projects "fun, affordable self-care"; email flows communicate "exclusive luxury." This is narrative fragmentation. It signals to the consumer that they cannot locate the brand's identity. Consistency correlates directly with trust and repeat purchase rates. 80% of consumers demand transparency and legible labeling. The most effective operator move is turning this demand into the narrative itself. Sourcing maps, ingredient origins, and clinical data become the story -- not just the footer content. For wellness brands under \$200M, the founder's voice remains the most potent vehicle for this consistency. → **Claims:** A claim made on a PDP must match the ad copy, which must match the package. Inconsistency triggers distrust and regulatory risk. → **Tone:** You cannot be "clinical and cold" on the site and "meme-centric" on Instagram without a bridging mechanism. → **Transparency:** If you claim "clean ingredients," the back-panel label must withstand scrutiny. A unified narrative lowers CAC because you aren't re-educating the customer on who you are at every touchpoint. *** ## Part 8: The Self-Assessment Checklist We quantify brand health to remove emotion from the roadmap. **The Scoring Rubric (1-5 Scale):** → **1: Non-existent.** No strategy or infrastructure; completely reactive. → **2: Reactive.** Activity exists, but ad-hoc; no intentional strategy. → **3: Functional.** Strategy exists; metrics tracked but unoptimized. → **4: Optimized.** Intentional strategy with measurable results; regularly reviewed. → **5: Frontier.** Best-in-class execution; compounds over time; defensive moat. | **Dimension** | **Score (1-5)** | | ------------------------------ | --------------- | | 1. Product Page Structure | \_\_\_ | | 2. Content Velocity | \_\_\_ | | 3. Email Capture Mechanics | \_\_\_ | | 4. Social Engagement Quality | \_\_\_ | | 5. Paid Media Efficiency | \_\_\_ | | 6. Brand Narrative Consistency | \_\_\_ | | **TOTAL** | **\_\_\_ / 30** | → **25-30 (Top 5%):** Market leaders. Focus on marginal gains. → **18-24 (Strong Foundation):** Profitable, but leaking efficiency. → **12-17 (Average):** Vulnerable to CAC spikes and competitor incursion. → **Under 12 (Urgent):** Existential risk. Unit economics likely broken. **Prioritization:** Fix the bucket before turning on the hose. 1. Lowest in Dim 1-2 → Fix the **Digital Shelf**. 2. Lowest in Dim 3-4 → Fix **Audience Infrastructure**. 3. Lowest in Dim 5-6 → Fix the **Efficiency Layer**. *** ## Part 9: Quick-Start Cheat Sheet One action per dimension this week: → **PDP:** Rewrite above-the-fold to answer *What is it?* and *Why me?* instantly. → **Content:** Batch produce 4 weeks of short-form video in one day. → **Email:** Change popup from generic discount to a specific product benefit. → **Social:** Reply to every comment on last 5 posts with a follow-up question. → **Paid:** Launch an Advantage+ campaign with 5 distinct creative angles. → **Narrative:** Audit your About page against your TikTok bio; align them. **Setup:** These prompts are designed for Gemini 3.1 Pro. Free via Google AI Studio ([aistudio.google.com](http://aistudio.google.com)). Start with Dimension 1 for fastest ROI. *** The gap between a \$10M brand and a \$50M brand is rarely product quality; it is the rigor of the systems surrounding the product. This is the exact audit framework we use internally at DAS. If your score reveals gaps you do not have the bandwidth to address -- that is exactly the work we do. # The Creative Velocity Playbook Source: https://intelligence.madebydas.com/playbooks/creative-velocity 17 workflows to out-supply the algorithm with distinct creative in 30-minute sprints. Claude AI plus Figma Console MCP for DTC visual production. > **Note:** This playbook is dense by design. It is meant to be built with, not skimmed once. Bookmark it, set it up, come back every time you need to ship faster. > > **Need the whole system built and run for you?** > > That is what DAS does -- signal detection, creative strategy, paid media execution, retention and lifecycle, and performance optimization as a *managed service*. We build the whole loop, running continuously, with a team that owns the outcomes, not the deliverables. > > If you need more than 20 distinct creative concepts a month and your current process can't deliver them, we should talk. [**amlan@madebydas.com**](mailto:amlan@madebydas.com) *** > **Note: Quick Start** > > The Stack, content, and workflows in this guide is relevant for any operator producing more than 20 visual assets a month: LinkedIn content creators, agency owners, founders building a personal brand, and marketing teams who are eager to increase output velocity. > > **For DTC and retail-brand operators:** Read Parts 1-4 in full. Start with Workflow 1 or 5. > > **For LinkedIn content creators and personal brands:** Jump to **Workflows 7, 11, 12, 13, and 15.** > > **For agency owners and brand managers:** Start with Workflows 2, 8, 14, and 17. > > **For those who are not in paid media:** Parts 1, 2, and 6 will permanently change how you think about visual production. > > **Everyone:** Part 5 prompts are copy-paste ready. You can use them today without the full Figma setup. *** ## Part 1: The Visual Production Bottleneck Most marketers are still using AI exclusively for copywriting. They write faster. They brief faster. They produce more words per hour than ever before. And then they wait three days for a designer to turn those words into something a human being actually wants to look at. That bottleneck is expensive, and it just got permanently removed. **The Math Was Already Broken** Brands are routinely signing \$50K+ agency retainers for basic design work. And those elite teams spend half their billed hours resizing ad variants, tweaking email headers, and swapping text on social cards. Premium talent should NOT be doing commodity labor. The platform data makes this worse. * Nielsen: Creative drives 56% of a campaign's sales ROI * Meta: Creative quality accounts for 50% of campaign effectiveness * LinkedIn: Carousels get 303% more engagement than standard text posts * Comments are weighted 8x heavier than likes in the current algorithm You cannot compete with limited visual assets anymore. You need visual velocity, and the old model can't produce it at the speeds the feeds demand. What would take a team of three now takes two operators + AI workflows -- a structural shift in how marketing teams can operate. **Why Creative Became the Primary Lever** Before you can understand why visual production volume matters more than it ever has, you must appreciate the eras of change: **Era 1: Broadcast.** One ad, mass audience. The constraint was cash; could you afford enough media to be seen? **Era 2: Internet Advertising.** Audience targeting arrived; behavioral data, demographics, interest pools, lookalikes. The constraint shifted to targeting skill -- could you find the right audiences before your competitors? **Era 3: AI-Powered Advertising (Now).** The platforms no longer need you to "define" audiences. They serve personalized creative to individual users based on predicted preferences. The constraint is two fold: creative supply, and knowing your audience -- the algorithm can only optimize from the options you give it. > If you can't out-target the algorithm anymore, you'll have to ***out-supply*** it. > > Different creative reaches different people. The variety of creative you produce ***is*** your targeting strategy. **What Does "Supply" Actually Mean in 2026?** This is where most brands get it wrong. They hear "produce more creative" and start generating 100 variations of the same ad -- same layout, swapped headline, different background color. Under [Meta's Andromeda retrieval system](https://searchengineland.com/meta-ai-driven-advertising-system-andromeda-gem-468020), those 100 variants register as a single entity in the auction, which means you're only getting one "ticket" to the lottery, which means you get one "chance" to win. Andromeda rewards genuinely different creative concepts. A UGC testimonial, a studio product demo, a text-overlay motion graphic, and a lifestyle Reel each register as a distinct entity and each gets its own chance in the auction. What matters is maximizing distinct creative concepts, not raw variant count. 15-25 truly different concepts will outperform 100 minor variations of three concepts every time. How many distinct concepts you actually need, guided by and scales with, what you're spending. Reference the breakdown below: | **Monthly Ad Spend** | **New Creatives Needed** | | -------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **\$10K-\$25K** | **3-4 per month** | | **\$25K-\$50K** | **1-2 per week** | | **\$50K-\$100K** | **3-5 per week** (this spend-range is when creative testing + iteration must be prioritized within your advertising) | | **\$100K-\$500K** | **15-20 per week** (this spend-range requires greater volume of creative testing + iteration; roughly 10-40% of total spend should be allocated for testing) | | **\$500K+** | **25-50% of budget allocated to testing** (At this range, it's very difficult to have both quality and volume in creative, and these workflows will ensure new ideas and iterations are always incoming, while having different creative strategists/creative teams involved to keep rotations fresh) | A healthy account running at scale needs to produce 20-100 genuinely distinct creative concepts per month. Each one needs to look, sound, and feel different from the others. Volume without variety is wasted production -- ***and that is the bottleneck.*** This guide solves for that. **The Three Traps Keeping Your Visual Pipeline Slow** **1. The Copy-Design Split:** Your content workflow has two tracks running in parallel, but they don't talk to each other. Copy is done in Claude or Google Docs. Design is done in Figma, Canva, or Photoshop. Every handoff between the two costs you a day. Multiply that by 50 assets a month and you've lost a week of production time that could have been eliminated entirely. **2. The Designer Constraint** Even the best marketing teams are throttled by design bandwidth. You need genuine creative diversity to feed the algorithm. Your designer can reasonably produce 20 distinct concepts per month, which means you run the same creative longer than you should, frequency climbs, performance degrades, and you blame the algorithm. The actual constraint was upstream, in the handoff, not the platform. **3. The Context Gap** When you brief a designer, you describe what you want. They interpret it. You revise. Two rounds of revision is the industry standard. That means every visual asset carries the cost of a misunderstanding. When Claude directs Figma directly, the brief and the execution happen in the same moment. The context gap closes. > **An Important Distinction That Matters** > > This is not about replacing creative directors, designers, or human taste. Your hero campaigns, your flagship product branding and positioning, your packaging: those still require a human mind, eyes, and direction. > > This is strictly about automating Tier 2 production: your daily ad variants, social cards, email headers, and carousel content. The high-volume, low-complexity assets that drain your budget and your team's time. > > You build the master template once with human taste. The machine scales it infinitely with **zero** marginal cost. *** ## Part 2: The Stack -- Claude AI + Figma Console MCP You need a brain and you need hands. Claude is the brain -- It writes the copy, analyzes the data, and makes layout decisions. Figma is the hands -- It executes the design, manipulates the pixels, and exports the files. The critical detail most people miss: **you cannot use the official Figma MCP for this.** The official Figma MCP is read-only. Read-only is useless for automated production. You need the **Figma Console MCP built by Southleft**. It is read-write. It actually builds and edits the files on your canvas. ### Tool 1: Claude AI (Intelligence Layer) Claude is your creative director, copywriter, and briefing engine in one. It takes your spoken or typed brief and converts it into precise, structured design instructions that Figma executes directly. Paired with **Wispr Flow** for voice input, you can speak a full design brief at 180 WPM instead of typing it at 40. The idea stays intact because you never lose it to keyboard fatigue. Paired with a **CLAUDE.md file**, your brand colors, type system, spacing rules, and tone load automatically into every session. Every output is on-brand by default; limited (or zero) briefing required. ### Tool 2: Figma Console MCP by Southleft (Execution Layer) The Southleft Figma Console MCP is a read-write Model Context Protocol server. It uses a secure WebSocket connection and a Desktop Bridge Plugin to give Claude direct programmatic access to your live Figma document. Claude is not generating a screenshot of a design. It is writing to your actual Figma file -- placing components, updating text nodes, swapping fills, and exporting production-ready assets. > **Note:** This stack was not chosen randomly; we **tested** the alternatives and they **broke** under pressure. **The Setup: 10 Minutes, One Time** You do not need to be a developer to configure this. You just need to follow instructions. 1. Install **Node.js 18 or higher** on your local machine. 2. Download and install the **Figma Desktop application** (not the browser version). 3. Open Figma settings and generate a new **Personal Access Token**. 4. Install your MCP client -- **Claude Desktop** is the recommended starting point. Claude Code and Cursor also work. 5. Clone the [**Southleft Figma Console MCP repository**](https://github.com/southleft/figma-console-mcp) from GitHub. 6. Open your terminal and run the standard install command from the repo readme. 7. Input your Figma Personal Access Token when the terminal prompts you. 8. Open Figma Desktop and install the **Desktop Bridge Plugin** from the Southleft repo. 9. Connect Claude Desktop to the local server port provided in your terminal. 10. Create a **CLAUDE.md file** in your project folder with your brand system. See Part 5, Prompt 7 for the exact template. Your engine is now live and ready to receive commands. *** ## Part 3: The 57 Tools -- Your AI Design Command Center The Southleft MCP gives Claude 57 specific capabilities inside Figma. Think of them as buttons on an industrial control panel. I have organized them into plain-English categories. You won't need to memorize these, but you need to know exactly what each tool does. **Navigation and Status** 1. Ping -- Checks if Figma is awake and actively listening to Claude. 2. Get Current File -- Tells Claude exactly which file you have open. 3. Get Selection -- Shows Claude which specific element you just clicked on. **Console Debugging** 1. Read Logs -- Pulls error codes if an automated build fails. 2. Clear Logs -- Wipes the slate clean for a fresh test. 3. Set Log Level -- Controls how much technical data Claude reports back to you. **Utility** 1. Export Node -- Saves your finished design to your desktop as a PNG or PDF. 2. Get Viewport -- Tells Claude exactly what part of the canvas you are looking at. **Real-Time Tracking** 1. Watch Node -- Monitors a specific design element for any manual changes. 2. Unwatch Node -- Stops monitoring that specific element. 3. Get Watchers -- Lists everything Claude is currently tracking in the background. **Design System Extraction** 1. Get Text Styles -- Pulls all your brand fonts into a usable list. 2. Get Color Styles -- Extracts your exact brand HEX codes. 3. Get Effect Styles -- Grabs your exact shadow and blur settings. 4. Get Grid Styles -- Maps out your specific layout columns. 5. Get Components -- Finds every master template hidden in your file. 6. Get Component Sets -- Groups your variant templates together logically. 7. Get Variables -- Pulls your strict spacing and sizing rules. 8. Get Variable Collections -- Organizes your rules by specific themes. 9. Export Design Tokens -- Packages your brand rules for web developers. **Design System Kit** 1. Create Text Style -- Builds a new reusable font rule for the team. 2. Create Color Style -- Saves a new brand color directly to your kit. 3. Create Effect Style -- Standardizes a new drop shadow across the file. 4. Create Grid Style -- Builds a new column layout template. 5. Publish Styles -- Pushes your new rules to the whole team library. **Design Creation** 1. Create Frame -- Draws a new blank artboard on the canvas. 2. Create Text -- Drops a new text box onto the frame. 3. Create Shape -- Draws exact rectangles, circles, or lines. **Component Management** 1. Create Component -- Turns a raw design into a master template. 2. Create Instance -- Drops a working copy of your template onto the canvas. 3. Update Component -- Changes the master template permanently. 4. Detach Instance -- Breaks the link between a copy and its master. **Variable Management** 1. Create Variable -- Sets a strict numerical rule for future designs. 2. Update Variable -- Changes that specific numerical rule. 3. Delete Variable -- Removes the rule entirely from the system. 4. Create Variable Collection -- Builds a dedicated folder for your rules. 5. Update Variable Collection -- Renames or moves the folder. 6. Delete Variable Collection -- Trashes the folder completely. 7. Bind Variable -- Forces a design element to obey your specific rule. 8. Unbind Variable -- Frees the element from the strict rule. 9. Get Bound Variables -- Shows which rules are currently controlling an element. 10. Set Variable Mode -- Switches a design instantly from light mode to dark mode. 11. Get Variable Modes -- Lists all available theme options in the file. **Node Manipulation** 1. Update Node -- Changes the physical size or position of an element. 2. Delete Node -- Deletes the element entirely from the canvas. 3. Group Nodes -- Bundles multiple loose items together. 4. Ungroup Nodes -- Breaks the bundle apart into individual pieces. 5. Set Auto Layout -- Forces elements to space themselves perfectly. 6. Remove Auto Layout -- Turns off the automatic spacing rules. 7. Set Text Content -- Injects your AI-written copy directly into the text box. 8. Set Fill -- Changes the background color or image of an element. 9. Set Stroke -- Adds a precise border to an element. **Comments** 1. Get Comments -- Reads feedback notes left by your team members. 2. Add Comment -- Drops a new sticky note directly on the design. 3. Resolve Comment -- Marks the team feedback as completed. **Design-Code Parity** 1. Get CSS -- Translates the visual design into clean website code. 2. Copy SVG Code -- Grabs the raw vector data for developer handoff. **The 7 Power Tools** Marketers do not use all 57 tools daily, and you will spend 90% of your time using just seven of them. | **Tool Name** | **What It Actually Does For You** | | ---------------- | ------------------------------------------------------------------------- | | Get Current File | Tells Claude exactly where to work so it doesn't edit the wrong document. | | Create Instance | Drops your approved master template onto the canvas for editing. | | Set Text Content | Injects your freshly generated AI copy into the design automatically. | | Update Node | Resizes the text box perfectly if the new copy is slightly longer. | | Set Fill | Swaps the background color or image instantly for rapid A/B testing. | | Export Node | Saves the final, polished asset directly to your desktop folder. | | Read Logs | Tells you exactly why a process failed so you can fix it immediately. | *** ## Part 4: 17 Visual Production Workflows These are the systems that print money for operators, and they work across your entire marketing funnel. **Workflows 1-10** are optimized for paid media and DTC operators. **Workflows 11-17** are optimized for LinkedIn content, personal brand, and organic social. **How to use these workflows** Each workflow is a pre-written voice command. You speak it into Wispr Flow -- or type it directly into Claude Cowork, and Claude executes it autonomously inside Figma. The "I" and "my" in each command are intentional. You are delegating to Claude in your voice, from your workspace. Commands referencing "my inbox folder" or "my CLAUDE.md" point to the folder structure you set up in Part 2. **NOTE:** If you haven't set up the Figma Console MCP yet, the prompts in Part 5 work standalone -- paste them into any Claude session and they'll produce copy and briefs you can hand off to a designer. *** ### 1. The Ad Batch Builder The most common production bottleneck in paid social is headline testing. You have five angles you want to try, but building five separate frames manually takes two hours and kills the momentum. This workflow builds all five in one pass. 1. Claude opens your campaign Figma file and creates a new page for the batch. 2. It builds five ad frames at the correct spec, applying your background color from CLAUDE.md. 3. It pulls headline variations from your inbox folder and injects one into each frame. 4. It places the product image, applies the brand CTA button component, and exports all five. ```text theme={null} "Open my Figma campaign file. Create a new page called 'Meta Q2 Batch.' Build 5 ad frames at 1080x1080. Use the dark navy background. For each frame, use a different headline from the list in my inbox folder. Place the product hero image in the top 60% of each frame. Use the orange CTA button component. Label each CTA: 'Shop Now.' Apply my brand type system throughout. Export all 5 as PNG at 2x. Save them named: Meta_Q2_[Headline#]_1080.png" ``` *** ### 2. The Carousel Card Factory LinkedIn carousels drive 303% more engagement than standard text posts. The format works beyond LinkedIn as well. The bottleneck is production -- building five consistent, on-brand cards manually takes longer than writing the content that goes in them. This workflow closes that gap. 1. Claude builds a five-card carousel structure with the correct frame dimensions. 2. It pulls your hook from your CLAUDE.md hook library for Card 1. 3. Cards 2-4 each receive one insight with a bold stat and supporting sentence. 4. Card 5 is built as a CTA card with a save prompt, then all five are exported. ```text theme={null} "Build a 5-card LinkedIn carousel in Figma. Frame size: 1080x1080. Card 1: Title card. Use Hook #3 from my CLAUDE.md hook library. Cards 2-4: Each gets one insight with a bold stat and one sentence. Card 5: CTA card. Orange background with white 'Save this' prompt. Apply my brand font system. Consistent padding: 64px all sides. Export all 5 as PNG 2x. Name them: Carousel_LI_[1-5]_1080.png" ``` *** ### 3. The A/B Variant Generator Your top performer is the best brief you have. The fastest way to find the next winner is to isolate one variable at a time and test it against the control. This workflow takes your current winner and produces four structured test variants -- one variable changed per frame, everything else held constant. 1. Claude locates your top-performing frame on the Winners page. 2. It creates a new test batch page and duplicates the frame four times. 3. Each variant changes exactly one element: headline, layout, background color, or CTA size. 4. All four variants plus the original control are exported and named for the media buyer. ```text theme={null} "Take the top-performing ad frame on my 'Winners' page in Figma. Create 4 variants on a new page called 'AB Test Batch.' Variant 1: Same layout, swap headline to [NEW HEADLINE]. Variant 2: Move headline below the image. Product image full bleed top. Variant 3: White background instead of dark. All other elements identical. Variant 4: Remove subhead entirely. Increase CTA button size by 20%. Export all 4 as PNG 2x." ``` ### 4. The Seasonal Refresh Sprint Every seasonal campaign requires the same work: pull up six templates, swap the background, update the copy, re-export everything. It is mindless labor at a designer's hourly rate. This workflow does all six in a single pass without touching the component structure. 1. Claude opens the Evergreen Templates page and identifies the six target frames. 2. It swaps every dark navy background to the new seasonal color using Set Fill. 3. It replaces season-specific copy across all frames using Set Text Content. 4. All six are exported at their original specs with the seasonal prefix applied. ```text theme={null} "Open my master brand asset file. Find the 'Evergreen Templates' page. I need a spring refresh on 6 templates: hero banner, email header, Instagram story, LinkedIn cover, product badge, and website popup. Swap the dark navy backgrounds to spring sage green — #7A9E7E. Replace any winter-specific copy with: 'Spring Collection Now Live.' Keep all component placements, logo locks, and type styles identical. Export all 6 at their required specs. Prefix: Spring2026_" ``` *** ### 5. The Multi-Format Launch Pack Every product launch needs the same five assets: feed post, story, paid ad, email header, site banner. Building them one at a time means inconsistent sizing, inconsistent copy, and a full day of production. This workflow builds the complete pack in one session. 1. Claude creates a new page in your Figma file dedicated to the launch. 2. It builds all five frames at the correct platform specs. 3. It applies the launch color system and injects the primary message and CTA across all formats. 4. All five are exported at their required specs and named by format. ```text theme={null} "I have a product launch on Friday. I need the full visual asset pack. Create a new page in my Figma file: 'Product Drop — [NAME].' Build: Instagram post (1080x1080), story (1080x1920), Meta feed ad (1200x628), email header (600x200), site banner (1440x160). Product name: [NAME]. Primary message: 'Now Available.' Use the launch color system from my CLAUDE.md: deep red on cream. Apply all brand components. Export all 5 at required specs." ``` *** ### 6. The Blog-to-Visual Converter A blog post with five strong statistics in it is already five Instagram posts. The content exists -- the conversion is the bottleneck. This workflow reads the document, extracts the most shareable data points, and builds the visual cards automatically. 1. Claude reads the blog post from your inbox folder. 2. It identifies the five most shareable statistics or insights. 3. It builds five quote cards in Figma with the stat in large display type, alternating brand backgrounds. 4. All five are exported and named for immediate posting. ```text theme={null} "Read the blog post saved in my inbox folder: 'trend-report-q2.md.' Extract the 5 most shareable statistics or insights from it. Build 5 Instagram quote cards in Figma, 1080x1080. Each card: one stat or insight in large display type, centered. Backgrounds alternate between our cream and our navy. Add logo mark bottom right on each. Export all 5 as PNG 2x. Prefix: IGQuote_TrendReport_" ``` *** ### 7. The Educational Carousel This is the standard LinkedIn PDF format. It drives the highest engagement of any post type on the platform. You use it to break down complex frameworks into bite-sized visual slides. The bottleneck is always the same: the thinking takes 20 minutes, the production takes two hours. This workflow collapses both. 1. Build a master carousel template component in Figma. 2. Feed Claude a raw, unformatted text post. 3. Claude splits the text logically across 8 slides. 4. Claude uses Create Instance to duplicate the master template 8 times. 5. Claude uses Set Text Content to inject the copy into each slide automatically. ```text theme={null} "Take this draft post. Use the Create Instance tool to generate 8 slides from my Carousel Master component. Inject the copy using Set Text Content. Ensure no text box overflows the safe zones. Export as a single PDF for LinkedIn upload." ``` *** ### 8. The Case Study Deck A case study has a universal structure: problem, solution, result, proof. Every client win you have can be formatted into four slides. The value is in the specificity of the outcome -- the production should be invisible. This workflow makes it invisible. 1. Claude reads your raw client notes and extracts the core narrative. 2. It maps the story into a strict four-slide framework. 3. It populates each slide of the Case Study Carousel component set. 4. The deck is exported as a PDF ready for LinkedIn or a sales conversation. ```text theme={null} "Review the raw client notes below. Extract the core narrative into a strict 4-slide visual framework. Slide 1: The problem they faced. Slide 2: The specific solution applied. Slide 3: The hard metrics achieved. Slide 4: A direct quote from the client. Populate the Case Study Carousel component set. Export as PDF." ``` *** ### 9. The Campaign Ad Variant Matrix Under Meta's Andromeda retrieval system, 30 variants of the same concept register as one entity in the auction. What you need is genuine creative diversity -- different formats, different hooks, different personas -- each registering as a distinct creative in the system. This workflow builds 30 genuinely different ads, not 30 reskins. 1. Claude builds a 3x5x2 matrix: three formats, five hooks, two personas. 2. For each cell, it generates the headline, visual direction, and copy block. 3. It builds each ad in Figma using the Master Ad component, one distinct concept per frame. 4. All 30 are exported and named by format, hook, and persona for the media buyer. ```text theme={null} "I need 30 ad variants for the [CAMPAIGN NAME] launch. Do not create 30 versions of the same ad. I need genuine creative diversity across three dimensions: Format: static image, text-overlay, UGC-style, carousel card. Hook: pain-point, aspiration, social proof, curiosity. Persona: first-time buyer, lapsed customer, loyal advocate. Build a matrix: 3 formats x 5 hooks x 2 personas. For each cell, give me the headline, visual direction, and copy block. Then build each ad in Figma using the Master Ad component. Export all 30 as PNG 2x. Name convention: Ad_[Format]_[Hook]_[Persona].png Save the full matrix brief as campaign-variant-matrix.md" ``` *** ### 10. The Performance Scorecard Visual Weekly creative performance data sits in a CSV. Making sense of it -- identifying what's working, what's leaking, what to kill -- requires turning numbers into a visual that a media buyer or creative director can read in 30 seconds. This workflow builds that visual directly from the data. 1. Claude reads your weekly performance CSV from the inbox folder. 2. It identifies the top three performers by CTR and the bottom two. 3. It builds a scorecard frame in Figma with a five-column layout and green/red indicators. 4. The scorecard is exported as a PNG ready for a Monday morning review. ```text theme={null} "Read the performance CSV in my inbox — this week's Meta creative data. Identify the top 3 performers by CTR and the bottom 2. Create a Figma frame at 1200x800 — a creative performance scorecard. Layout: 5 columns. Each: creative label, CTR, CPM, green/red indicator. Apply my data visualization style: dark background, data in white and orange. Title: 'Creative Performance — Week of [DATE].' Export as PNG 2x. Save as: Creative_Scorecard_[DATE].png" ``` *** ### 11. The Lead Magnet Cover Builder A lead magnet without a polished cover does not convert -- the cover is the product in the moment of the click. Building one manually in Figma, adding a 3D mockup, exporting at spec, takes 45 minutes minimum. This workflow does it in under five. 1. Claude updates the title and subtitle text nodes on your Lead Magnet Cover frame. 2. It exports the flat cover as a high-resolution PNG. 3. It applies that PNG as the fill for your 3D Mockup component. 4. The final 3D mockup is exported and named for immediate use. ```text theme={null} "Update the text nodes on the Lead Magnet Cover frame with: Title: [TITLE]. Subtitle: [SUBTITLE]. Export the node as a high-resolution PNG. Apply that exported PNG as the fill for the 3D Mockup component. Export the final 3D mockup as PNG 2x. Name: LeadMagnet_Cover_[TITLE]_3D.png" ``` *** ### 12. The Profile/Page Banner Refresher Your LinkedIn banner is the first brand impression on your profile. Most people set it once and forget it. The operators who use it as a rotating campaign asset -- updating it for every launch, every content push, every new offer -- get meaningfully more profile-to-connection conversion. This workflow makes the update take less than five minutes. 1. Claude selects the Company Banner frame in your Figma file. 2. It updates the headline text to the new campaign message. 3. It swaps the background fill to the campaign color variable. 4. It exports at the exact 1584x396 LinkedIn specification and names the file. ```text theme={null} "Select the Company Banner frame. Update the main headline text to: [NEW CAMPAIGN HEADLINE]. Change the background fill variable to: [CAMPAIGN COLOR VARIABLE]. Export the frame as PNG at the exact 1584x396 dimensions needed for LinkedIn. Name: Banner_[CAMPAIGN NAME]_[DATE].png" ``` *** ### 13. The Comment Magnet Batch High-contrast stat cards are the most reliable comment-driving format on LinkedIn. One strong statistic, white text on dark background, portrait format. The problem is producing enough of them consistently without burning creative time on commodity work. This workflow produces five in one pass -- a full month of this content type. 1. Claude creates five instances of your Stat Magnet component in Figma. 2. It injects one statistic per instance using Set Text Content. 3. It checks contrast and frame dimensions across all five. 4. All five are exported at portrait format and named sequentially. ```text theme={null} "Create 5 instances of the Stat Magnet component. For each instance, inject one statistic from the provided list into the main text node. Frame size: 1080x1350px. High contrast: white text on dark background. Statistics to inject: [LIST OF 5 STATS]. Export all 5 nodes as PNG 2x. Name: StatMagnet_[Stat#]_[DATE].png" ``` *** ### 14. The Before/After Transformation Card Before/after is the oldest persuasion structure in marketing because it works. A specific metric on the left, a specific outcome on the right, clear contrast between the two states. The difficulty is building the card cleanly so the contrast reads instantly. This workflow handles the production so you can focus on finding the right numbers. 1. Claude selects the Split Screen Transformation frame in your Figma file. 2. It injects the Before and After metrics into the correct text nodes. 3. It updates the caption labels for context on both sides. 4. It checks contrast ratios and exports at 2x. ```text theme={null} "Select the Split Screen Transformation frame. Inject the Before metric into the left text node: [BEFORE METRIC]. Inject the After metric into the right text node: [AFTER METRIC]. Update the Before label caption to: [BEFORE CONTEXT]. Update the After label caption to: [AFTER CONTEXT]. Check contrast on both sides to ensure readability. Export the frame as PNG 2x. Name: BeforeAfter_[SUBJECT]_[DATE].png" ``` *** ### 15. The Process/Framework Diagram Every operator has a proprietary process. Most of them live in a Google Doc or a slide deck that never gets seen. Turning a five-step framework into a visual diagram is the fastest way to make that thinking visible and shareable on LinkedIn. This workflow builds the diagram from the framework steps you provide. 1. Claude creates five instances of your Process Node component. 2. It injects the step name and description into each node using Set Text Content. 3. It arranges all five inside the Auto-Layout Container and connects them with directional arrows. 4. The diagram is exported as a PNG ready for posting. ```text theme={null} "Use the Process Node component. Create 5 instances. Fill them with the 5 steps of my framework: Step 1: [STEP NAME + DESCRIPTION] Step 2: [STEP NAME + DESCRIPTION] Step 3: [STEP NAME + DESCRIPTION] Step 4: [STEP NAME + DESCRIPTION] Step 5: [STEP NAME + DESCRIPTION] Ensure they sit cleanly inside the Auto-Layout Container frame. Connect the nodes with directional arrows. Export as PNG 2x. Name: Framework_[NAME]_[DATE].png" ``` *** ### 16. The Data Visualization Post Data posts outperform opinion posts by a wide margin on LinkedIn because they give people something concrete to share. The bottleneck is always the same: turning raw numbers into a clean visual takes design skill most operators don't have time to deploy. This workflow handles the chart build automatically. 1. Claude analyzes the dataset you provide and identifies the most surprising insight. 2. It determines the correct chart type and updates the bar heights or data points in the Figma chart frame. 3. It writes a five-word headline stating the insight plainly. 4. It adds the methodology caption and exports the frame. ```text theme={null} "Take this dataset: [PASTE DATA]. Update the heights of the 4 rectangles in the Bar Chart frame to represent the exact percentages calculated from this data. Update the data labels beneath each bar. Write a 5-word headline that states the single most surprising insight. Add the methodology caption: [SOURCE + DATE]. Export the frame as PNG 2x. Name: DataViz_[TOPIC]_[DATE].png" ``` *** ### 17. The Brand Kit Updater When a brand color or typeface changes, every template in your Figma file needs to update. Done manually, this is a find-and-replace nightmare that takes hours and misses things. Done through Variables, it takes two minutes -- because every element that references the variable updates automatically when the variable changes. 1. Claude updates the Primary and Secondary color variables to the new HEX values. 2. It updates the Heading and Body font variables. 3. Changes propagate instantly across every frame and component that references those variables. 4. Claude reports how many elements were updated so you can verify the scope. ```text theme={null} "Update the Primary Brand Color variable to: [NEW HEX]. Update the Secondary Color variable to: [NEW HEX]. Update the Heading Font variable to: [NEW FONT]. Update the Body Font variable to: [NEW FONT]. Apply these changes across all Variable Collections. Confirm how many frames and components were updated." ``` *** ## Part 5: 11 Copy-Paste Claude Prompts Do not write prompts from scratch. Use these exact inputs to drive the engine. Copy them, paste them, and execute immediately. ### Prompt 1: The Brand DNA Loader ```text theme={null} You are operating as my AI design director inside Figma. My brand system is as follows: Primary color: [HEX]. Secondary: [HEX]. Accent: [HEX]. Background colors: [list]. Text colors: [list]. Primary typeface: [FONT]. Headings: [WEIGHT/SIZE]. Body: [WEIGHT/SIZE]. Logo: located in my Figma component library under 'Brand / Logo.' Button style: use the 'CTA Button / Primary' component from the library. Spacing: 8px base unit. Standard padding: 48px or 64px for ads. Tone: [e.g., 'confident, direct, premium']. Every design you create must use this system. Do not improvise brand elements. Confirm you have loaded this before accepting any design task. ``` ### Prompt 2: The Ad Frame Builder ```text theme={null} Create a new Figma frame for a [PLATFORM] static ad. Dimensions: [WIDTHxHEIGHT]. Campaign objective: [OBJECTIVE]. Target audience: [DESCRIPTION]. Headline: [HEADLINE TEXT]. Subhead (optional): [SUBHEAD TEXT]. CTA: [CTA TEXT] — use the primary CTA button component. Product image: [FILE PATH or 'use the hero from the component library']. Apply my full brand system. Prioritize [ELEMENT] as the dominant visual. Export as PNG 2x. Name: [PLATFORM]_[CAMPAIGN]_[DATE].png ``` *** ### Prompt 3: The Variant Generator ```text theme={null} Take the frame named [FRAME NAME] on the [PAGE NAME] page. This is my current top performer. I need 4 test variants. Create a new page: '[CAMPAIGN] — Test Batch [DATE].' Variant A: Change only the headline to: [NEW HEADLINE]. Variant B: Change only the CTA copy to: [NEW CTA]. Variant C: Swap background color from [CURRENT] to [TEST COLOR]. Variant D: [LAYOUT CHANGE]. All other elements remain identical to the control. Export all 4 plus the original control as PNG 2x. Name: [CAMPAIGN]_Control.png, [CAMPAIGN]_VarA.png, etc. ``` *** ### Prompt 4: The Carousel Outline Generator ```text theme={null} Act as a direct-response copywriter. Take the topic provided below. Break it into exactly 8 slide concepts for a LinkedIn carousel. Slide 1 must be a hook with a bold claim. Slides 2-7 must each cover one actionable step. Slide 8 must be a specific call to action. Do not write final copy yet. Just give me the structural outline. ``` *** ### Prompt 5: The Slide Copy Injector ```text theme={null} Take the 8-slide outline we just agreed on. Write exact micro-copy for each slide. Keep sentences strictly under 8 words. No filler. Format as a strict JSON object. Key each slide by number (e.g., "slide_1"). Include keys for "headline", "body_text", and "image_prompt". This JSON will be injected directly into Figma via Set Text Content. ``` *** ### Prompt 6: The Multi-Format Launch Pack Brief ```text theme={null} I have a [LAUNCH TYPE] launching on [DATE]. Build a complete visual asset pack on a new Figma page: '[LAUNCH NAME] Assets.' Required formats: Instagram feed (1080x1080), Instagram story (1080x1920), Meta feed ad (1200x628), email header (600x200), site banner (1440x160). Master message: [PRIMARY HEADLINE]. Secondary line: [SUBHEAD]. CTA across all formats: [CTA TEXT]. Apply my full brand system. Export all at 2x PNG. Name: [LAUNCH]_[FORMAT]_[SPEC].png ``` *** ### Prompt 7: The CLAUDE.md Design Block Template ```text theme={null} ## DESIGN SYSTEM — AUTO-LOAD FOR ALL FIGMA SESSIONS Primary Brand Color: [HEX] Secondary: [HEX] | Accent: [HEX] | Background: [HEX] | Dark: [HEX] Typography: Display: [FONT] [WEIGHT] — headlines and hero text Body: [FONT] [WEIGHT] — subheads, body copy Labels: [FONT] [WEIGHT] — CTAs, captions, metadata Spacing: 8px base unit. Ad padding: 48px mobile / 64px desktop. Figma Component Library Location: [PROJECT NAME / PAGE NAME] CTA Button: 'Buttons / Primary CTA' Logo: 'Brand / Wordmark — [COLOR VARIANT]' Product Card: 'Cards / Product — Standard' Export Rules: Always PNG at 2x unless otherwise specified. File Naming: [BRAND]_[FORMAT]_[VARIANT]_[DATE] DO NOT improvise colors or type outside this system. DO NOT delete any existing layers — always work on duplicates. DO NOT use stock or placeholder images — use brand assets only. ``` *** ### Prompt 8: The Brand Kit Definer ```text theme={null} I am building a brand kit for [COMPANY TYPE] targeting [AUDIENCE]. The tone is [TONE DESCRIPTION]. Give me a strict visual rulebook. Provide exactly 3 HEX codes (Primary, Secondary, Accent). Provide 2 Google Fonts (Heading, Body). Provide a strict spacing scale in pixels (e.g., 4, 8, 16, 32). Output this as a list of Figma Variables to create, formatted for direct input into the Variable Management tools. ``` *** ### Prompt 9: The Data Visualization Planner ```text theme={null} Analyze the dataset provided below. Identify the single most surprising or counterintuitive insight. Tell me exactly which type of chart to build in Figma (Bar, Line, or Donut). Write a 5-word headline that states the insight plainly — no jargon. Write a 2-sentence caption explaining the methodology and data source. Do not editorialize. State the data. Let the numbers make the argument. ``` *** ### Prompt 10: The Banner Copy Variations Generator ```text theme={null} My current offer is [DESCRIBE YOUR OFFER]. I need 5 variations of copy for my LinkedIn profile banner. Each variation must include: - A main headline strictly under 6 words. - A sub-headline stating the specific outcome the client gets. - A call to action pointing to the link in my bio. Focus entirely on time saved and capital preserved. Output as a numbered list, ready to test as banner variants in Figma. ``` *** ### Prompt 11: The Campaign Ad Copy Generator ```text theme={null} Review the campaign brief below. Generate 10 distinct ad copy variations. Matrix them across 2 angles: [ANGLE 1] and [ANGLE 2]. For each variation, provide: - A 4-word image overlay headline. - A 2-sentence primary text caption. Output as a JSON array formatted for immediate Figma injection. Each object should include keys: "overlay_headline", "primary_text", "angle". Campaign brief: [PASTE BRIEF] ``` *** ## Part 6: The 20-Minute Weekly System This is not a solo grind. It is a repeatable system any operator can run. You do not need to be a trained designer to execute this. You just need strict discipline and a calendar. *** ### Monday: Signal to Brief (5 minutes) Pull your weekly performance data from Meta or your analytics tool. Identify your top-performing creative format from last week. Note one format to scale and one angle to test. Load your Brand DNA Loader prompt. Write three lines: what you're making, what message it carries, what it needs to do. That is your brief. ### Tuesday-Wednesday: Production (10 minutes) Use Prompt 2 to generate the primary asset for the week's campaign. Use Prompt 3 on your existing top performer to produce 3-4 test variants. If you have a launch this week, run Prompt 6 while you're already in the session. Export everything. ### Thursday: Review + Deploy (5 minutes) Run the 3-second scroll test. Check text legibility at mobile size. Upload to your ad manager or hand to your media buyer. Tag each creative by format, angle, and campaign. ### Friday: Audit (passive) Monitor comments. Note which formats are pulling engagement. Feed the data back into next Monday's brief. Four weeks of this consistency will completely rewire your content operation. Four weeks of this consistency will completely rewire your account. *** ### The 5 Rules for Visual Delegation **1. Always load Brand DNA first.** Every session without it produces off-brand output. **2. One variable per variant.** Test headline OR layout OR color -- never two at once. **3. Use real component names, not descriptions.** "The Primary CTA button component" -- not "an orange button with rounded corners." **4. Name files on export, not after.** Build your naming convention into every prompt. **5. Run 5 assets per session, then stop.** After 5 complex design tasks, context gets crowded. Fresh session for the next batch. *** ## Part 7: Visual Specs Cheat Sheet ### Meta | Format | Dimensions | Notes | | ----------------- | -------------- | ----------------------------------- | | Feed -- Square | 1080 x 1080 px | Primary format for most campaigns | | Feed -- Landscape | 1200 x 628 px | Link previews and awareness | | Feed -- Portrait | 1080 x 1350 px | Max real estate in the feed | | Story / Reels | 1080 x 1920 px | Keep top and bottom 250px clear | | Carousel Card | 1080 x 1080 px | Must be consistent across all cards | ### LinkedIn | Format | Dimensions | Notes | | --------------------- | -------------- | --------------------------------- | | Carousel / PDF Slide | 1080 x 1350 px | 4:5 portrait gets max feed height | | Single Image Post | 1200 x 628 px | Standard feed format | | Single Image Portrait | 1080 x 1350 px | Higher engagement format | | Profile Banner | 1584 x 396 px | Update for every new campaign | | Company Cover | 1128 x 191 px | Center-weighted safe zone | | Profile Photo | 400 x 400 px | -- | | Company Logo | 300 x 300 px | -- | ### TikTok | Format | Dimensions | Notes | | ---------- | -------------- | ------------------------------- | | In-Feed Ad | 1080 x 1920 px | Avoid bottom 20% -- UI overlaps | | End Card | 1080 x 1920 px | Center 60% for key message | ### Email | Format | Dimensions | Notes | | ------------- | ------------ | ---------------------- | | Header Banner | 600 x 200 px | Left-align key content | | Content Image | 600 x 400 px | Standard margins | ### Website | Format | Dimensions | Notes | | --------------------- | ------------- | ---------------------------- | | Hero Banner (Desktop) | 1440 x 560 px | Center-safe zone: 800px wide | | Site Popup | 600 x 500 px | Keep CTA above fold | ### Display Ads | Format | Dimensions | Notes | | ---------------- | ------------ | ----------------------------- | | Leaderboard | 728 x 90 px | Minimal text, clear CTA | | Medium Rectangle | 300 x 250 px | Brand mark top-left or right | | Half Page | 300 x 600 px | Strong visual, single message | ### Export Rules * LinkedIn carousels: PDF, max 100MB, 8-12 slides * LinkedIn single images: PNG, max 5MB * Meta ads: PNG, max 30MB * Email images: PNG, max 200KB * Never use JPG for text-heavy graphics * Always at 2x for retina clarity * Min font size for social legibility: 32px * Min contrast ratio: 4.5:1 for all text on backgrounds * Safe margin for PDF carousels: 60px minimum ### Pre-Deploy Checklist 1. File size under platform limit? 2. Primary text readable at mobile size? 3. Correct file format exported? 4. CTA visible and clear? 5. Brand components used -- no improvised elements? If yes on all five, **deploy**. *** ## Recap You have the complete framework now to run a visual production engine without a design bottleneck. You have the stack: Claude AI and the Southleft Figma Console MCP + the exact 10-step setup. You have the 57 tools mapped. You have **17 copy-paste workflows** covering every visual production scenario: paid media, product launches, seasonal refreshes, LinkedIn carousels, personal brand assets, data visualization, lead magnets, and more. You have **11 prompts** ready to deploy today. And you have the 20-minute weekly sprint that keeps your pipeline full without hiring, briefing, or waiting. The visual bottleneck was never about creativity. It was about handoffs. We just fixed that. > **Note:** **Need the whole system built and run for you?** > > This is what DAS does -- signal detection, creative strategy, paid media execution, retention and lifecycle, and performance optimization as a *managed service*. We build the whole loop, running continuously, with a team that owns the outcomes that you will be proud of. > > If you need more than 20 distinct creative concepts a month and your current process can't deliver them, we should talk. [**amlan@madebydas.com**](mailto:amlan@madebydas.com) # The Growth Loop Playbook: 5 Claude Skills for Brand Strategy Source: https://intelligence.madebydas.com/playbooks/growth-loop 5 purpose-built Claude Skills for the ecommerce Growth Operating Loop — consumer intelligence, trend detection, competitive tracking, campaign briefing, and performance synthesis. 5 Purpose-Built Claude Skills for the Ecommerce Growth Operating Loop *** Notion_Cover_Square.png Your growth team runs five workflows every week — review mining, trend detection, competitive tracking, campaign briefing, performance synthesis. These five workflows consume 60–80% of your strategic bandwidth. And at most brands, every single one of them is manual. This playbook gives you a Claude Skill for each stage of the loop. Install them. Run the loop. Let the efficiencies compound. *** # Quick Start: Download These Skills For those who prefer diving into the deep end, you can get started by downloading the skills. Each file is a .md skill you upload to Claude, and they begin to work immediately; no setup, no code, no API keys. For best results, also upload a Brand Brain document (template included below). ### Download All 5 Skills | Skill | What It Does | | -------------------------------- | ------------------------------------------------------------------------------------------- | | **Consumer Intelligence Engine** | Mines reviews and feedback for messaging insights, ad copy angles, and objection frameworks | | **Trend Radar** | Detects and scores emerging category signals across 6 platforms | | **Competitive Position Tracker** | Analyzes competitor pricing, creative, messaging, and finds whitespace | | **Campaign Brief Architect** | Generates complete omnichannel creative briefs from your intelligence | | **Performance Command Center** | Turns weekly metrics into narrative intelligence and prioritized action lists | *** ### 3-step install 1. **Download** the .zip file(s) above 2. **Upload** to Claude: go to Customize → Skills → upload the .zip 3. **Ask Claude anything relevant** — the skill activates automatically That is it. Claude detects when a skill applies and loads it behind the scenes. No slash commands, no prompting tricks. You describe what you need, the skill brings the methodology. > **Want these skills customized for your brand — with your competitors, your products, your channel mix pre-loaded?** That is what DAS builds. > > Reach out: [amlan@madebydas.com](mailto:amlan@madebydas.com) *** # The Problem Nobody's Solving Every scaling ecommerce brand runs the following five workflows: * They mine customer reviews for messaging insights. * Review-mining means exporting CSVs from your primary online storefront, social storefronts, marketplaces, retail, affiliate, etc., then pasting them into spreadsheets, and spending hours tagging themes by hand. * They scan the market for emerging trends. * Trend detection means scrolling TikTok and hoping something clicks or scouring Google trends and Reddit for insights. * They track what competitors are doing. * Competitive intelligence means checking the Meta Ad Library when someone remembers to. * They brief creative for campaigns. * Campaign briefing means a Slack thread with a few bullet points that the creative team has to interpret. * And they synthesize performance data into decisions about what to do next. * Meaning, pulling reports from six dashboards, copy-pasting into a deck, and presenting numbers without clear next steps. These five workflows consume 60–80% of a growth team's strategic bandwidth. And at most brands, each workflow is happening manually. The cost is not measured in dollars. It is measured in speed. **The brands winning right now have compressed this loop from weeks to hours.** They are not smarter. They are not better funded. They have simply automated the connective tissue between insight and action. The signal that used to take two weeks to travel from a customer review to an ad-headline now takes an afternoon. The tools exist. 89% of retail and ecommerce companies are testing AI in some capacity. But only 7–26% have piloted deployments. The gap between belief and execution is enormous—and it is almost entirely an implementation problem, not a technology problem. The barrier is not skepticism. 98% of mid-market marketers believe AI will improve their effectiveness. The barriers are skill gaps, integration challenges, and not knowing where to start. Generic AI tools produce generic output. And the ecommerce-specific AI tooling that exists is either prohibitively expensive or disconnected from the workflows that actually drive growth. This playbook closes that gap with five purpose-built Claude Skills—one for each stage of the Growth Operating Loop—that a growth team can install and start using today. *** # The Growth Operating Loop The five workflows are not independent tasks. They are stages in a continuous cycle that compounds over time. We call this the Growth Operating Loop. The_Growth_Loop_Diagram.png Each stage maps to one skill: 1. **Understand → Consumer Intelligence Engine.** Mine reviews, support tickets, and social conversations to extract what customers actually say, want, and object to. 2. **Detect → Trend Radar.** Identify category signals accelerating across platforms before competitors see them. Classify by durability. Score by opportunity. 3. **Position → Competitive Position Tracker.** Monitor competitor pricing, ad creative, messaging, and product launches. Map where you stand. Find the whitespace. 4. **Execute → Campaign Brief Architect.** Transform insights from the first three skills into structured, channel-specific creative briefs your team can immediately act on. 5. **Measure → Performance Command Center.** Synthesize cross-channel performance data into narrative intelligence that closes the gap between what happened and what to do next. **The compounding effect is the point; each cycle through the loop makes the next cycle sharper:** → Consumer intelligence informs trend detection → Trend detection shapes competitive positioning → Competitive positioning drives campaign briefs → Campaign performance feeds back into consumer intelligence A brand running this loop manually completes it once per quarter, maybe. A brand running it with these five skills completes it weekly. Over a year, that is 52 cycles of compounding intelligence versus 4. The gap in market responsiveness becomes insurmountable. The rest of this playbook gives you everything you need to install and run the full loop. *** # What Are Claude Skills (And Why They Matter) If you have used Claude before, you have used prompts; the one-off instructions that disappear when the conversation ends. Every new session, you start from zero. You re-explain your brand, re-describe your audience, re-state your formatting preferences. Skill_Constellation_Diagram.png Claude Skills change this entirely. **A Skill is a *persistent*, reusable set of instructions** that Claude loads automatically whenever a relevant task is detected. Think of it as a trained specialist you hire once. | | Prompt | Skill | | --------------- | --------------------------- | ------------------------- | | Persistence | Gone after conversation | Permanent until removed | | Context | You provide every time | Pre-loaded automatically | | Triggering | You type the instruction | Claude detects relevance | | Output quality | Varies with prompt quality | Consistent methodology | | Brand awareness | Only what you paste in | Reads from Brand Brain | | Reusability | Copy-paste between sessions | Install once, use forever | ### How to Install a Skill 1. **Enable Skills.** In Claude, go to Settings → Capabilities → toggle on Code execution and file creation. Then go to Customize → Skills. 2. **Upload the Skill file.** Download the .zip from the Quick Start section above. Upload it in Customize → Skills. 3. **Verify activation.** Ask Claude: "What skills are available?" You should see the skill listed. 4. **Upload your Brand Brain.** For best results, also upload the Brand Brain document (template below) as a Project file or directly in your conversation. Skills work across Claude.ai (web and mobile), Claude Code (command line), and Claude Cowork (autonomous agent mode). Available on Free, Pro, Max, Team, and Enterprise plans. *** # Setting Up Your Brand Brain Every skill in this playbook references a shared Brand Brain—a single context document that captures everything Claude needs to produce output that sounds like your brand, understands your market, and speaks to your specific customer. **Without a Brand Brain, these skills produce good output. With a Brand Brain, they produce YOUR output.** ### The Essential Sections Your Brand Brain does not need to be exhaustive on day one. Start with these six sections and expand over time. **1. Brand Voice and Tone** How your brand sounds. Include 3–5 adjectives that describe your tone, examples of on-brand and off-brand language, words or phrases you always or never use, and reading level. **2. Target Personas** Go beyond demographics. Include psychographic drivers: why they buy, what they fear, what outcome they are really purchasing, what language they use to describe their problem (pull directly from Skill 1 outputs), and what objections they raise. **3. Product Catalog** Core products with key proof points. For each: one-sentence positioning, primary and secondary benefits, price point, top customer-cited reasons for purchasing, and compliance guardrails. **4. Competitive Set** Top 3–5 direct competitors: their positioning, pricing relative to yours, messaging themes, and specific reasons a customer chooses you over them (and vice versa). Feeds directly into Skill 3. **5. Channel Mix and Priorities** Primary sales channels (Shopify, Amazon, Walmart, retail), marketing channels ranked by investment, current budget allocation, and KPI targets by channel. Feeds Skills 4 and 5. **6. Key Metrics and Targets** Revenue targets, MER target, CAC ceiling, ROAS thresholds by channel, conversion rate benchmarks, email/SMS revenue contribution targets. *** > **Building a comprehensive Brand Brain takes weeks of focused work to do well.** DAS builds custom Brand Brains for scaling ecommerce brands — including competitive research, persona development, and channel strategy mapping — as part of our AI workflow implementation engagements. > > Reach out: [amlan@madebydas.com](mailto:amlan@madebydas.com) *** # Skill 1: Consumer Intelligence Engine *Understand what your customers actually say, want, and object to.* Skill1_Consumer_Intelligence_Diagram.png *** **↑ *`Download consumer-intelligence-engine.md file`*** *** ## The Problem Your brand has thousands of reviews scattered across Shopify, Amazon, Trustpilot, Reddit, and social media. Individually, each review is anecdotal. Together, they contain your entire messaging strategy, written in the exact language your customers use. The problem is synthesis. Most brands either ignore this data entirely or run it through generic sentiment tools that classify everything as "positive" or "negative" without extracting anything actionable. A skincare brand scraping 10,000 Amazon reviews found their AI tools frequently misclassified sentiment because they did not understand terms like "texture," "absorption," or "glow." The output said 72% positive. The team still had no idea what to do. Practitioners who do this well report 40–70% conversion rate increases simply by replacing internal marketing language with phrases extracted directly from customer conversations. One wellness brand discovered users described their pre-product state as "feeling scattered and overwhelmed" and their desired outcome as "finally being able to take a full breath." After rebuilding messaging around these exact phrases, trial-to-paid conversion jumped 32% and CAC dropped nearly 50%. The words your customers use to describe their problem are your headline copy. This skill extracts them systematically. ## The Workflow The Consumer Intelligence Engine runs a four-layer analysis on any set of customer reviews or feedback data: ```text theme={null} ┌──────────────────────────────────────────────────────────────┐ │ CONSUMER INTELLIGENCE ENGINE │ │ │ │ INPUT: Reviews, support tickets, social feedback │ │ ───────────────────────────────────────────── │ │ │ │ Layer 1 ─── Aspect-Level Sentiment │ │ (packaging: neg | taste: pos | price: neutral) │ │ │ │ │ Layer 2 ─── Theme Extraction │ │ (2-3 verbatim quotes per theme) │ │ │ │ │ Layer 3 ─── Competitive Benchmarking │ │ (side-by-side sentiment vs. competitors) │ │ │ │ │ Layer 4 ─── Actionable Output Generation │ │ 6 deliverables (see below) │ │ │ │ ───────────────────────────────────────────── │ │ OUTPUT: │ │ ✓ Messaging Matrix ✓ Ad Copy Angle Library │ │ ✓ Product Feedback List ✓ FAQ Content Map │ │ ✓ Objection Framework ✓ Competitive Voice Comparison │ └──────────────────────────────────────────────────────────────┘ ``` * **Layer 1 — Aspect-Level Sentiment.** Classifies sentiment at the feature level (packaging: negative; taste: positive; price: neutral) rather than just overall polarity. This surfaces the specific product dimensions driving satisfaction and dissatisfaction. * **Layer 2 — Theme Extraction.** Identifies recurring themes using topic modeling with frequency analysis. Extracts 2–3 direct verbatim quotes per theme to preserve authentic customer language. Groups themes by: pain points described, desired outcomes, product-specific feedback, and purchase triggers. * **Layer 3 — Competitive Benchmarking.** Compares sentiment patterns against your top competitors (using reviews from the same platforms). Identifies where you outperform, where you underperform, and where customers are making direct comparisons. * **Layer 4 — Actionable Output Generation.** Produces six deliverables: a messaging matrix mapping customer language to marketing copy angles, an ad copy angle library organized by theme, a product feedback priority list ranked by frequency multiplied by severity, an FAQ content map, an objection handling framework, and a competitive voice comparison. ## The SKILL.md Below is the complete skill code. Placeholders marked with `[CUSTOMIZE]` should be replaced with your brand-specific information, ideally pulled from your Brand Brain document. The skill is designed to work immediately with reasonable defaults, but customization dramatically improves output quality. ```markdown theme={null} --- name: consumer-intelligence-engine description: > Analyzes customer reviews, support tickets, and social feedback to extract messaging insights, sentiment patterns, and voice-of-customer intelligence for ecommerce brands. Use when asked to analyze reviews, mine customer language, build messaging matrices, or extract VOC data. --- # Consumer Intelligence Engine ## Context You are a senior consumer research analyst specializing in ecommerce and retail brands. You extract actionable messaging intelligence from customer feedback data — not summaries, not sentiment scores, but the specific language, themes, and patterns that drive conversion. Read the Brand Brain document for brand context, competitive set, and target personas before beginning analysis. ## Input Handling Accept review data in any format: - CSV exports from Shopify (Yotpo, Judge.me, Okendo, Stamped) - Amazon review dumps (from Helium 10, Jungle Scout, or manual copy) - Pasted text from Trustpilot, Reddit, social comments - Support ticket exports from Gorgias, Zendesk, or Kustomer If data includes star ratings, use them. If not, infer sentiment from language. Flag when sample size is below 50 reviews per product. ## Analysis Framework ### Layer 1: Aspect-Level Sentiment For each product or product category in the data: - Identify the top 8-12 aspects customers mention (ex: taste, texture, packaging, price, effectiveness, shipping, scent) - Classify each mention as positive, negative, or neutral - Calculate aspect-level sentiment ratios - Flag aspects with >30% negative sentiment as priority issues - [CUSTOMIZE: Add your product-specific aspects here] ### Layer 2: Theme Extraction Group feedback into themes. For each theme: - Name the theme in plain language (not jargon) - Provide frequency count or proportional indicator - Extract 2-3 VERBATIM quotes that best represent the theme - Classify as: Pain Point | Desired Outcome | Product Feedback | Purchase Trigger | Objection | Competitor Comparison - NEVER fabricate quotes or frequency counts. Use ranges if unsure. ### Layer 3: Competitive Benchmarking If competitor review data is provided or referenced: - Compare aspect-level sentiment across brands - Identify aspects where you outperform competitors - Identify aspects where competitors outperform you - Note where customers make explicit brand comparisons - [CUSTOMIZE: Your competitive set from Brand Brain] ### Layer 4: Deliverables Generate ALL of the following: **1. Messaging Matrix** | Customer Language | Pain Point | Desired Outcome | Copy Angle | Map exact customer phrases to the marketing message they suggest. **2. Ad Copy Angle Library** Organize verbatim phrases by theme. Each angle includes: the customer quote, the implied benefit, and a suggested headline. **3. Product Feedback Priorities** Ranked list: frequency x sentiment severity. Top 5 issues with specific customer quotes and recommended actions. **4. FAQ Content Map** Questions customers are asking (explicitly or implicitly) mapped to recommended answers using voice-of-customer language. **5. Objection Handling Framework** Top 5-8 objections with: the objection in customer language, the underlying concern, the counter-argument, and proof points. **6. Competitive Voice Comparison** (if data available) Side-by-side: how your customers describe you vs. how competitor customers describe them. Focus on language differences. ## Output Format Structure output with clear headers for each deliverable. Use tables where they aid readability. Always lead with the Messaging Matrix — it is the highest-value output. Include a 3-sentence executive summary at the top. ## Guardrails - Never hallucinate frequency data. If counts are uncertain, say so. - Never invent quotes. Every quoted phrase must appear in the source. - Flag sample sizes below statistical significance. - Note if data skews toward one sentiment (e.g., only 1-star reviews). - [CUSTOMIZE: Add any brand-specific compliance guardrails] ``` ## See It Work: Sample Output Here is what the Messaging Matrix output looks like for a hypothetical wellness supplement brand analyzing 2,400 Amazon reviews: > **EXECUTIVE SUMMARY:** Customers consistently describe their pre-purchase state using language centered on energy and mental fog, not ingredient-specific terms. The strongest purchase trigger is peer recommendation from specific communities (fitness, new mothers). The primary objection is price relative to grocery-store alternatives. > > **MESSAGING MATRIX (Top 5 of 18 angles identified)** > > ```markdown theme={null} > ┌────────────────────────────────────┬──────────────────────┬────────────────────────┬───────────────────────────────┐ > │ Customer Language │ Pain Point │ Desired Outcome │ Copy Angle │ > ├────────────────────────────────────┼──────────────────────┼────────────────────────┼───────────────────────────────┤ > │ "I was tired of feeling foggy │ Afternoon energy │ Sustained mental │ "The 2pm fog, gone." │ > │ by 2pm" │ crash │ clarity │ │ > ├────────────────────────────────────┼──────────────────────┼────────────────────────┼───────────────────────────────┤ > │ "My trainer recommended it and │ Skepticism from past │ Credible recommendation│ "The supplement your trainer │ > │ I actually noticed a difference" │ supplements │ + felt result │ actually uses." │ > ├────────────────────────────────────┼──────────────────────┼────────────────────────┼───────────────────────────────┤ > │ "It's expensive but I stopped │ Cost justification │ Consolidation / │ "Replace your cabinet. │ > │ buying three other things" │ │ simplification │ One product." │ > ├────────────────────────────────────┼──────────────────────┼────────────────────────┼───────────────────────────────┤ > │ "Finally something that doesn't │ Taste/texture │ Enjoyable daily │ "Wellness that doesn't taste │ > │ taste like chalk" │ aversion │ routine │ like a punishment." │ > ├────────────────────────────────────┼──────────────────────┼────────────────────────┼───────────────────────────────┤ > │ "I was skeptical but my wife │ Trust barrier │ Social validation │ "Don't take our word for it. │ > │ convinced me to try it" │ │ │ Take hers." │ > └────────────────────────────────────┴──────────────────────┴────────────────────────┴───────────────────────────────┘ > ``` > > *\[...continues for all 18 angles across 6 deliverables]* ## Customization Notes * **Vertical adaptation:** Replace aspect categories for your product type. A fashion brand analyzes fit, fabric quality, and sizing accuracy. A pet brand analyzes palatability, ingredient sourcing, and pet reaction. The analysis framework is identical — only the aspects change. * **Data source tips:** Amazon reviews are highest volume. Shopify reviews tend to skew more positive (customers already committed). Reddit and social comments contain the rawest, most unfiltered language. The best analyses combine all three. * **Frequency:** Run this monthly at minimum. Run it immediately after a product launch, a major promotion, or a negative PR event. The messaging matrix should be a living document your content team references weekly. > **Want this built for your brand?** > > DAS builds custom Consumer Intelligence Engines with pre-mapped aspect taxonomies for your product category, automated data ingestion templates, and integration with your review management platform. → [amlan@madebydas.com](mailto:amlan@madebydas.com) *** # Skill 2: Trend Radar *Spot what is accelerating in your category before your competitors do.* Skill2_Trend_Radar_Diagram.png *** ***↑ Download trend-radar.md file*** *** ## The Problem Trend forecasting for ecommerce brands operates on two tracks: 1. Long-range planning (12–24 months) draws on trend-forecasting agencies like [WGSN](https://www.wgsn.com/en) and [Mintel](https://www.mintel.com/) for macro cultural shifts. 2. Near-term intelligence (6–12 months) requires real-time signal tools like [Spate](https://www.spate.nyc/), which analyzes 900 billion Google search signals and 200 million social posts with 72% accuracy on 12-month predictions. Most brands aren’t actively investing in either track. So they rely on gut feel, whatever the founder saw on TikTok last weekend, or trend reports that are already six months stale by the time they reach the team. The result is predictable. Brands either chase trends too late (the "everyone is doing it" phase where there is no differentiation) or invest in trends too early without validation (the "we were ahead of our time" phase that burns budget). The brands that appear prescient are not clairvoyant. They have a signal detection system that distinguishes velocity from noise. This skill gives you that system. It does not replace a WGSN subscription for 24-month macro planning. It replaces the ad hoc, manual scanning that most teams do for 1–12 month planning — which is where the majority of campaign and content decisions actually happen. ## The Workflow ```text theme={null} ┌──────────────────────────────────────────────────────────────┐ │ TREND RADAR │ │ │ │ INPUT: Google Trends, TikTok, Reddit, Amazon, │ │ Pinterest, social listening data │ │ ───────────────────────────────────────────── │ │ │ │ Step 1 ─── Signal Detection │ │ (identify accelerating patterns) │ │ │ │ │ Step 2 ─── Cross-Platform Validation │ │ 1 platform = noise │ 3+ = validated │ │ │ │ │ Step 3 ─── Opportunity Scoring │ │ Magnitude × Momentum × Monetization │ │ (1-5 each, max composite: 125) │ │ │ │ │ Step 4 ─── Trend Classification │ │ Megatrend│Macrotrend│Microtrend│Fad │ │ │ │ │ Step 5 ─── Risk Assessment + Action Recommendation │ │ │ │ ───────────────────────────────────────────── │ │ OUTPUT: Scored + classified Trend Brief │ │ with specific action recs per signal │ └──────────────────────────────────────────────────────────────┘ ``` * **Signal Detection.** Accepts data from Google Trends, TikTok Creative Center, Reddit threads, Amazon search term reports, Pinterest Trends, and social listening exports. Identifies signals that are accelerating — not just popular, but gaining speed. * **Cross-Platform Validation.** A signal appearing on one platform is noise. A signal appearing on three platforms is a pattern. The skill cross-references signals across sources and assigns a validation score based on platform diversity. * **Opportunity Scoring.** Each validated signal is scored on three dimensions: Magnitude (how many consumers care deeply), Momentum (growth rate and trajectory), and Monetization (willingness to pay a premium). These multiply into a composite opportunity score. * **Trend Classification.** Signals are classified into four tiers: Megatrend (10+ years, foundational bets), Macrotrend (3–10 years, R\&D and product line investments), Microtrend (6 months to 3 years, content and limited-edition plays), and Fad (days to weeks, social content only — never invest in product development). * **Action Recommendation.** Each classified trend receives a specific recommended action: ignore, create content, launch limited-edition SKU, brief R\&D, or adjust positioning. ## The SKILL.md ```markdown theme={null} --- name: trend-radar description: > Detects and validates emerging trends in ecommerce categories by analyzing signals across Google Trends, TikTok, Reddit, Amazon, Pinterest, and social platforms. Classifies trends by durability and scores them by commercial opportunity. Use when asked about trends, emerging categories, what is gaining traction, or market signals. --- # Trend Radar ## Context You are a trend intelligence analyst for ecommerce brands. Your job is not to identify what is popular — it is to identify what is ACCELERATING and assess whether it represents a durable commercial opportunity or a temporary spike. Read the Brand Brain for: product category, competitive set, current positioning, and the specific sub-categories or ingredients relevant to this brand. ## Input Handling Accept trend data in any combination of: - Google Trends exports or described patterns - TikTok Creative Center data or described observations - Reddit thread summaries or links - Amazon Search Term reports (from Brand Analytics or Helium 10) - Pinterest Trends data or Pinterest Predicts references - Social listening exports (Brandwatch, Sprout, Meltwater) - Freeform descriptions of observed signals If web search is available, supplement provided data with current search to validate signals. If not, work with provided data only and note where additional validation would strengthen conclusions. ## Analysis Framework ### Step 1: Signal Extraction From provided data, identify distinct signals. A signal is: - A specific topic, ingredient, format, aesthetic, or behavior - That shows measurable acceleration (not just presence) - Within the brand's category or adjacent categories For each signal, note: source platform, acceleration evidence, approximate timeline of emergence, and consumer segment driving it. - [CUSTOMIZE: Your category keywords and sub-categories] ### Step 2: Cross-Platform Validation For each signal, check (or ask) if it appears across platforms: - 1 platform = Unvalidated (note, do not act) - 2 platforms = Emerging (monitor weekly) - 3+ platforms = Validated (evaluate for action) Assign validation score: Low / Medium / High ### Step 3: Opportunity Scoring Score each validated signal on three dimensions (1-5 scale): MAGNITUDE: How many consumers care? 1 = Niche community (<10K active participants) 3 = Growing segment (100K-1M active participants) 5 = Mass market (1M+ active participants) MOMENTUM: How fast is it growing? 1 = Stable/flat interest 3 = 20-50% growth in search/mention volume over 90 days 5 = 100%+ growth, viral acceleration MONETIZATION: Will they pay? 1 = Interest only, no purchase intent signals 3 = Active product searching, comparison shopping 5 = Existing premium pricing, strong willingness to pay COMPOSITE SCORE = Magnitude x Momentum x Monetization (max 125) ### Step 4: Trend Classification Based on evidence pattern, classify each signal: MEGATREND (10+ years) — Foundational shift in consumer behavior. Action: Inform long-term product strategy and brand positioning. Example: Personalized nutrition, AI-assisted shopping. MACROTREND (3-10 years) — Established and growing category shift. Action: R&D investment, product line extension, hiring. Example: Functional mushrooms in wellness, clean beauty. MICROTREND (6 months - 3 years) — Specific and actionable now. Action: Content plays, limited-edition SKUs, campaign angles. Example: Specific ingredient spikes, aesthetic movements. FAD (days - weeks) — Viral but not durable. Action: Social content ONLY. Never invest in product development. Example: Viral TikTok challenges, meme-driven interest spikes. ### Step 5: Risk Assessment For each high-scoring signal, note potential risks: - Regulatory exposure (FDA, FTC claims) - Supply chain constraints - Backlash potential (cultural sensitivity, greenwashing) - Competitive saturation risk (how many brands already here?) - [CUSTOMIZE: Industry-specific risk factors] ## Output Format Structure as a TREND BRIEF with: 1. Executive Summary (3 sentences: biggest opportunity, biggest risk, recommended immediate action) 2. Trend Dashboard table: | Signal | Platforms | Validation | M x M x M | Class | Action | 3. Deep Dive on top 3 signals (one paragraph each with evidence) 4. Whitespace Opportunities (where NO competitor is playing yet) 5. Watch List (unvalidated signals to monitor next cycle) ## Guardrails - Never conflate popularity with acceleration. - Never classify a fad as a macrotrend to sound optimistic. - Always note when data is insufficient for confident classification. - Include at least one contrarian call: a popular signal you believe is DECLINING or at risk of backlash. ``` ## See It Work: Sample Output Here is a condensed Trend Brief output for a functional beverage brand analyzing Q1 2026 signals: > **EXECUTIVE SUMMARY:** Magnesium-based relaxation products show the strongest acceleration this quarter (composite score: 100/125) with validated signals across TikTok, Reddit, Google Search, and Amazon. The "cortisol conscious" movement represents a macrotrend reframing stress management from mental health to hormonal optimization. Recommend: launch a magnesium SKU brief within 30 days and create 5 content pieces around cortisol education this week. > > **TREND DASHBOARD (Top 5 of 12 signals analyzed)** > > ```markdown theme={null} > ┌─────────────────────┬───────────┬────────────┬───────┬───────────┬──────────────────┐ > │ Signal │ Platforms │ Validation │ Score │ Class │ Action │ > ├─────────────────────┼───────────┼────────────┼───────┼───────────┼──────────────────┤ > │ Magnesium drinks │ 4 │ High │ 100 │ Macrotrend│ Product brief │ > │ Cortisol-conscious │ 3 │ High │ 75 │ Macrotrend│ Content series │ > │ Mushroom coffee │ 2 │ Medium │ 30 │ Fading │ Reduce investment│ > │ "Protein water" │ 3 │ High │ 60 │ Microtrend│ Limited SKU test │ > │ "Sleepy girl │ 1 │ Low │ 15 │ Fad │ Social only │ > │ mocktail" │ │ │ │ │ │ > └─────────────────────┴───────────┴────────────┴───────┴───────────┴──────────────────┘ > ``` > > **CONTRARIAN CALL:** Mushroom coffee shows declining search velocity despite continued shelf space expansion. Early adopters have moved on. Brands still investing in this category face margin compression within 6 months. ## Customization Notes * **Vertical adaptation:** For fashion brands, replace ingredient signals with aesthetic and material signals (ex: "quiet luxury" declining, "demure maximalism" accelerating). For home goods, track material trends, color palette shifts, and design movement signals. The scoring framework is universal. * **Data sourcing tips:** Google Trends is your baseline (free, quantitative). Reddit is your leading indicator (signals appear here 3–6 months before mainstream). TikTok shows velocity but inflates magnitude. Amazon Search Terms from Brand Analytics show commercial intent, which is the strongest monetization signal. * **Frequency:** Run monthly for ongoing category intelligence. Run immediately before product planning cycles, seasonal content planning, or investor presentations. > **Want this built for your brand?** DAS builds custom Trend Radars with automated signal feeds for your category, competitive tracking overlays, and integration with your product development calendar. → [amlan@madebydas.com](mailto:amlan@madebydas.com) *** # Skill 3: Competitive Position Tracker *Know where you stand, what they are doing, and where the whitespace is.* Skill3_Competitive_Position_Diagram.png *** ***↑ Download competitive-position-tracker.md file*** *** ## The Problem Nearly 83% of ecommerce companies still track competitors using manual spreadsheet-based methods. 44% report having zero competitor visibility beyond sporadic checks. Meanwhile, the competitive intelligence software market is projected to reach \$1.12 billion by 2032—**evidence that brands know this matters, even if most have not solved it.** The current tool landscape is fragmented and expensive. Pricing intelligence tools like Prisync cover price monitoring but miss messaging. Ad creative tools like Foreplay archive competitor ads but do not connect creative strategy to market positioning. Full-stack CI platforms like Klue are built for enterprise B2B, not ecommerce growth teams. The result is that competitive intelligence at most brands is reactive and incomplete. Someone checks the Meta Ad Library before a campaign launch. Someone notices a competitor's price change on Amazon. But no one is systematically connecting pricing shifts to messaging changes to product launches to ad creative strategy. That connection is where the real intelligence lives. This skill creates a structured, repeatable competitive analysis workflow that connects all four dimensions and outputs actionable positioning intelligence. ## The Workflow ```text theme={null} ┌──────────────────────────────────────────────────────────────┐ │ COMPETITIVE POSITION TRACKER │ │ │ │ INPUT: Ad Library screenshots, pricing data, competitor │ │ websites, email captures, Amazon listings │ │ ───────────────────────────────────────────── │ │ │ │ Dimension 1 ─── Ad Creative Analysis │ │ (theme, hook, visual, copy, format, intent) │ │ │ │ Dimension 2 ─── Pricing Intelligence │ │ (effective price incl. bundles/subs/shipping)│ │ │ │ Dimension 3 ─── Messaging & Positioning Shifts │ │ (website, email, ads, social evolution) │ │ │ │ Dimension 4 ─── Whitespace Identification │ │ (what NO competitor is doing) │ │ │ │ ───────────────────────────────────────────── │ │ OUTPUT: │ │ ✓ Competitive Dashboard ✓ Creative Strategy Brief │ │ ✓ Pricing Position Map ✓ Battlecard per competitor │ │ ✓ Whitespace Opportunities (ranked) │ └──────────────────────────────────────────────────────────────┘ ``` * **Ad Creative Analysis.** Analyzes competitor ad creative across six dimensions: main theme/angle, hook or opening, visual treatment, copy and CTA, format and structure, and overall strategic intent. Identifies what they are running longest (their winners) and what they are testing (their hypotheses). * **Pricing Intelligence.** Tracks effective pricing including bundles, subscriptions, shipping thresholds, and promotional cadence. Calculates effective unit cost, not just sticker price. Maps pricing relative to your position. * **Messaging and Positioning Shifts.** Monitors changes in competitor website copy, email subject lines, ad messaging, and social content to detect strategic repositioning. Flags when a competitor shifts target audience, value proposition, or brand tone. * **Whitespace Identification.** Analyzes the full competitive landscape to identify positioning, messaging, and creative angles that no competitor is currently occupying. These are your highest-opportunity campaign starting points. ## The SKILL.md ```markdown theme={null} --- name: competitive-position-tracker description: > Analyzes competitor pricing, ad creative, messaging, and positioning for ecommerce brands. Produces competitive battlecards, positioning maps, and whitespace analysis. Use when asked about competitors, market positioning, competitive landscape, or pricing intelligence. --- # Competitive Position Tracker ## Context You are a competitive intelligence analyst for ecommerce and retail brands. You analyze competitors not to copy them, but to find the gaps they leave open and the positioning they are conceding. Read the Brand Brain for: competitive set, current pricing, positioning, and channel strategy. ## Input Handling Accept competitive data in any combination of: - Meta Ad Library screenshots or described observations - Competitor website URLs or described changes - Pricing screenshots or reported price points - Competitor email subject lines or campaign descriptions - Ad creative images or described creative approaches - Amazon listing data (pricing, A+ content, BSR) - Social media posts or content strategy observations - [CUSTOMIZE: Your primary competitors by name] ## Analysis Framework ### Dimension 1: Ad Creative Analysis For each competitor's ad creative, analyze across six axes: 1. MAIN THEME: What is the primary angle? (Price, quality, social proof, lifestyle aspiration, problem-solution, ingredient story) 2. HOOK: What stops the scroll? (Question, stat, bold claim, UGC testimonial, visual contrast, before/after) 3. VISUAL TREATMENT: Production level, color palette, photography vs illustration, text overlay approach, brand consistency 4. COPY & CTA: Tone, length, urgency mechanisms, offer structure 5. FORMAT: Static, carousel, video, UGC, comparison, listicle 6. STRATEGIC INTENT: Prospecting vs retargeting, brand vs direct response, top-of-funnel vs bottom-of-funnel Identify: longest-running ads (their proven winners), newest ads (their current hypotheses), and format distribution patterns. ### Dimension 2: Pricing Intelligence For each competitor, calculate EFFECTIVE price: - Sticker price - Subscription discount (if offered) - Bundle savings (per-unit in bundle vs standalone) - Shipping threshold and effective shipping cost - Current promotions and promotional frequency - [CUSTOMIZE: Your pricing structure for comparison] Map pricing positions relative to your brand: Premium (>20% above you) | Comparable | Value (<20% below you) ### Dimension 3: Messaging & Positioning Shifts Track changes in competitor messaging across channels: - Website headline and hero copy evolution - Email subject line patterns and themes - Ad messaging themes over time - Social content themes and tone shifts Flag significant shifts: new target audience signals, value prop changes, brand tone adjustments, new claim types. ### Dimension 4: Whitespace Analysis Based on full competitive landscape, identify: - Messaging angles NO competitor is using - Audience segments NO competitor is targeting explicitly - Content formats NO competitor has adopted - Pricing positions that are unoccupied - Channel strategies competitors are ignoring ## Output Format Generate ALL of the following: **1. Competitive Dashboard** Summary table: each competitor's current positioning, pricing tier, primary ad strategy, and biggest vulnerability. **2. Creative Strategy Brief** What competitor ads are running longest and why. What you should NOT copy (saturated) and what angles remain open. **3. Pricing Position Map** Visual description of where each competitor sits on price vs perceived value. Identify your optimal position. **4. Battlecard** For each competitor: their pitch, your counter, proof points, and the specific customer objection that favors you. **5. Whitespace Opportunities** Ranked list of unoccupied positions with: the opportunity, why it is open, risk of pursuing it, and recommended first move. ## Guardrails - Analyze strategy, not just tactics. Why are they doing this? - Never recommend copying competitors. Recommend countering them. - Note data freshness — competitive data degrades quickly. - Flag when competitive data is incomplete or single-source. - [CUSTOMIZE: Competitor-specific context from Brand Brain] ``` ## See It Work: Sample Output Here is a condensed Battlecard output for one competitor of a hypothetical DTC skincare brand: > **COMPETITOR: GlowBase** > Positioning: Clinical efficacy at accessible pricing > Pricing: \$38 hero SKU (\$32 on subscription) vs. your \$45 (\$38 sub) > Primary Ad Strategy: Before/after UGC with dermatologist endorsement > Running Longest: "28-day challenge" carousel (12+ weeks active) > > ```markdown theme={null} > ┌─────────────────────────────────────────────────────────────┐ > │ BATTLECARD: GlowBase │ > ├──────────────┬──────────────────────────────────────────────┤ > │ Their Pitch │ "Dermatologist-developed formulas that work │ > │ │ in 28 days" │ > ├──────────────┼──────────────────────────────────────────────┤ > │ Your Counter │ "Clean ingredients that work without the │ > │ │ clinical compromise" │ > ├──────────────┼──────────────────────────────────────────────┤ > │ Proof Points │ Full INCI list on homepage vs their │ > │ │ "proprietary blend" | 4.7 avg on 8,200 │ > │ │ reviews vs their 4.3 on 3,100 │ > ├──────────────┼──────────────────────────────────────────────┤ > │ Objection │ "I want effective skincare but I don't trust │ > │ That Favors │ brands that hide behind 'proprietary' │ > │ You │ formulas" │ > ├──────────────┼──────────────────────────────────────────────┤ > │ Vulnerability│ No email/SMS presence visible. Entire │ > │ │ strategy appears Meta-dependent. │ > └──────────────┴──────────────────────────────────────────────┘ > ``` > > **WHITESPACE:** No competitor in this set is running comparison-style content ("us vs. them" ingredient breakdowns). This format has high engagement in adjacent categories and is completely uncontested here. ## Customization Notes * **Vertical adaptation:** For fashion, emphasize aesthetic positioning and influencer strategy over ingredient analysis. For food and beverage, add retail shelf positioning and distributor strategy. For supplements, add regulatory claim tracking as a fifth dimension. * **Data sourcing tips:** Meta Ad Library is free and the single best source for competitor ad creative. Use Foreplay or a simple screenshot workflow to archive what you see. For pricing, check competitor websites on a set schedule — tools like Prisync automate this, but a weekly manual check works for small competitive sets. * **Frequency:** Run the full analysis monthly. Run the ad creative scan biweekly. Run an emergency scan whenever a competitor launches a new product, changes pricing, or enters a new channel. > **Want this built for your brand?** DAS builds automated competitive monitoring dashboards with bi-weekly battlecard updates, pricing alert triggers, and integration with your campaign planning workflow. → [amlan@madebydas.com](mailto:amlan@madebydas.com) *** # Skill 4: Campaign Brief Architect *Turn intelligence into structured, channel-specific creative briefs your team can immediately act on.* Skill4_Campaign_Brief_Diagram.png *** ***↑ Download campaign-brief-architect.md file*** *** ## The Problem Creative production is the binding constraint on growth for most ecommerce brands. A 10-SKU brand selling across Amazon, Walmart, and social ads needs 2,500+ assets per year minimum. E-commerce creative fatigues in 7–14 days on Meta versus 4–8 weeks for B2B. But the bottleneck is rarely creative talent. It is the brief. Unclear or incomplete briefs are the number one cause of wasted creative cycles. "Make a cool ad" is not a brief. "We need more sales" is not a strategy. Vague requests lead to endless revisions, wasted ad spend, and campaigns that miss the mark. Teams spend more time going back and forth on what they want than it takes to actually produce the asset. The other failure mode is channel fragmentation. A single campaign idea needs to be adapted for Meta (1:1, 4:5, 9:16), TikTok (9:16 native aesthetic), Email (mobile-first with subject line variants), Amazon A+ (5–8 image sizes per SKU), and potentially influencer briefs with usage rights specifications. Most brands either produce generic one-size-fits-all assets or manually adapt each one — both are expensive in different ways. **This skill ingests outputs from Skills 1–3 plus performance data from Skill 5 and generates complete, structured, channel-specific campaign briefs that a creative team can immediately execute against.** ## The Workflow ```text theme={null} ┌──────────────────────────────────────────────────────────────┐ │ CAMPAIGN BRIEF ARCHITECT │ │ │ │ INPUT: Campaign objective + outputs from Skills 1-3 & 5 │ │ ───────────────────────────────────────────── │ │ │ │ Feeds from other skills: │ │ ├── Skill 1: Messaging angles, customer language │ │ ├── Skill 2: Trend signals, category opportunities │ │ ├── Skill 3: Competitive whitespace, counter-positioning │ │ └── Skill 5: Past performance, winning formats │ │ │ │ │ Generates 13-section brief with: │ │ ├── Campaign foundation (objective, audience, message) │ │ ├── Channel-specific deliverable specs │ │ │ ├── Meta Ads (static, video, carousel specs) │ │ │ ├── TikTok (native format, sound, trend integration) │ │ │ ├── Email / SMS (subject lines, segmentation) │ │ │ ├── Amazon A+ (module layout, image specs) │ │ │ └── Influencer (creator criteria, usage rights) │ │ └── Execution framework (timeline, budget, success metrics) │ │ │ │ ───────────────────────────────────────────── │ │ OUTPUT: Complete brief executable WITHOUT a kickoff meeting │ └──────────────────────────────────────────────────────────────┘ ``` * **Campaign foundation:** Campaign name, project owner (single accountable person), objective tied to one KPI, background and "why now" context, target audience with psychographic depth pulled from Consumer Intelligence Engine output. * **Creative direction:** Core message (one takeaway), proof points and reasons to believe, creative mandatories and brand guardrails, reference assets and swipe file notes. Pulls competitive whitespace from Skill 3 to identify differentiated angles. * **Channel specifications:** Separate deliverable lists for each active channel with exact format specs, aspect ratios, character counts, and platform-native requirements. Includes Meta, TikTok, Email/SMS, Amazon A+, and Influencer brief sections. * **Execution framework:** Timeline with milestones, production and media budget allocation, success metrics with specific targets, and stakeholder approval chain. * **Performance integration:** When Skill 5 data is available, the brief incorporates what worked previously — top-performing formats, winning hooks, and creative elements to carry forward or retire. ## The SKILL.md ```markdown theme={null} --- name: campaign-brief-architect description: > Generates structured, channel-specific campaign briefs for ecommerce brands. Produces complete creative briefs with deliverable specs for Meta, TikTok, Email/SMS, Amazon, and Influencer channels. Use when asked to brief a campaign, plan creative, or generate a content brief. --- # Campaign Brief Architect ## Context You are a senior campaign strategist who writes briefs that creative teams love. Your briefs are specific enough to execute against without a follow-up meeting, yet strategic enough to give the creative team room for great work. Read the Brand Brain for: voice, personas, product info, channel mix, and brand guardrails. If outputs from other Growth Loop skills are available (Consumer Intelligence, Trend Radar, Competitive Position Tracker, Performance Command Center), incorporate their insights directly. ## Input Handling Accept campaign requests in any format: - High-level campaign concept or objective - Product launch details - Seasonal or promotional calendar - Performance data suggesting creative refresh needed - Outputs from other Growth Loop skills - Freeform description of what is needed When input is vague, ask clarifying questions for: objective/KPI, target audience, budget range, and timeline. Do not generate a brief without at least an objective and audience. ## Brief Structure Generate a complete brief with ALL 13 sections: ### 1. Campaign Overview - Campaign name (suggest if not provided) - Project owner: [CUSTOMIZE: default team lead] - Launch date and duration - One-sentence campaign summary ### 2. Objective - ONE primary KPI (ex:'Achieve \$25 target CPA on Meta for new customer acquisition' NOT 'increase brand awareness and drive sales') - Secondary KPI (optional, max one) ### 3. Background & Why Now - What triggered this campaign? - Market context (from Trend Radar if available) - Competitive context (from Position Tracker if available) ### 4. Target Audience - Primary persona (from Brand Brain) - Psychographic drivers: why they buy, what they fear, what language they use (from Consumer Intelligence Engine if available) - Exclusions: who this is NOT for ### 5. Core Message - ONE takeaway the audience should remember - Supporting messages (max 3) ### 6. Proof Points - Statistics, testimonials, certifications, awards - Customer quotes (from Consumer Intelligence Engine if available) ### 7. Creative Mandatories - Brand elements required (logo, colors, fonts, disclaimers) - Legal/compliance requirements - [CUSTOMIZE: Your brand guidelines] ### 8. Channel Deliverables For EACH active channel, provide a separate section: **Meta Ads:** - Formats: Static (1:1, 4:5), Video (9:16, 4:5), Carousel - Hook requirement: First 3 seconds must stop scroll - Variants: Min 3 headline variants for Advantage+ testing - UGC-style variant required for broad targeting **TikTok:** - Format: 9:16 vertical, native aesthetic (not polished ad) - Sound: On by default, trending audio when relevant - Hook: First 1.5 seconds decisive - Trend integration notes **Email / SMS:** - Subject line variants (min 3) with preview text - Hero image specs - CTA hierarchy (primary and secondary) - Segmentation criteria - Mobile-first design requirement **Amazon A+ Content:** (if applicable) - Module layout recommendation - Image sizes required - Enhanced Brand Content text - [CUSTOMIZE: Your Amazon brand requirements] **Influencer:** (if applicable) - Creator selection criteria - Content usage rights and duration - FTC disclosure requirements - Performance benchmarks and payment structure ### 9. Reference Assets - Internal past performers to reference (from Performance data) - Competitive creative to counter (from Position Tracker) - Mood/style references ### 10. Timeline - Brief approval date - Creative first draft due - Review rounds (max 2 recommended) - Final assets due - Launch date ### 11. Budget - Production budget - Media budget by channel (if known) ### 12. Success Metrics - Primary KPI target (specific number) - Secondary metrics to monitor - Measurement window - [CUSTOMIZE: Your standard KPI targets] ### 13. Approval Chain - Who reviews, who approves, who has final say - [CUSTOMIZE: Your approval workflow] ## Output Format Present as a structured document with clear headers. Each channel section should be independently readable. Include specific deliverable counts and dimensions. The brief should be actionable WITHOUT a kickoff meeting. ## Guardrails - Never produce a brief with more than one primary KPI. - Never produce a brief without channel-specific specs. - Always include a timeline, even if estimated. - Flag when input is insufficient for a complete brief. - [CUSTOMIZE: Brand-specific creative guardrails] ``` ## Example Output Here is the Meta Ads section from a campaign brief generated for a hypothetical supplement brand launching a new magnesium product (informed by Trend Radar output): > **CHANNEL: META ADS** > Objective: Achieve \$28 CPA for new customer acquisition over 30-day launch window > > **DELIVERABLES:** > > * 3x static images (1:1 and 4:5 aspect ratios) — product hero with benefit headline > * 2x short-form video (9:16 and 4:5) — 15 seconds max, UGC-style testimonial format > * 1x carousel (4:5) — 5 cards: hook stat, problem, solution, proof, CTA > * 3x headline variants for Advantage+ testing > > **HOOK STRATEGY (informed by Consumer Intelligence Engine):** > > * Angle A: "The 2pm fog, gone." (energy crash pain point — highest frequency in review data) > * Angle B: "The supplement your trainer actually uses." (peer credibility — strongest purchase trigger) > * Angle C: "Replace your cabinet. One product." (consolidation — strongest price objection counter) > > **COMPETITIVE DIFFERENTIATION (informed by Position Tracker):** > > * No competitor running comparison-style content. Test ingredient transparency vs. "proprietary blend" competitors. > * GlowBase's 28-day challenge format is proven but saturated. Avoid this format. Counter with immediate-result framing. > > **CREATIVE MANDATORIES:** Logo in lower-right, brand purple palette, no health claims without asterisk disclaimer, all lifestyle imagery must feature 30-45 age range. ## Customization Notes * **Channel adaptation:** Remove or add channel sections based on your actual mix. A brand that does not sell on Amazon drops that section entirely. A brand heavy on influencer marketing expands that section with detailed creator criteria. * **Integration with other skills:** This skill produces the best output when it has access to outputs from Skills 1–3. Feed it your latest messaging matrix, trend brief, and competitive battlecard, and the resulting campaign brief will be specific, differentiated, and grounded in real data rather than assumptions. * **Frequency:** Generate a new brief for every campaign, product launch, or seasonal push. Use the Performance Command Center output (Skill 5) to inform the brief with what worked and what did not in the previous cycle. > **Want this built for your brand?** DAS builds fully integrated Campaign Brief Architects that auto-pull from your Brand Brain, consumer intelligence, competitive tracking, and performance data — generating launch-ready briefs in minutes. → [amlan@madebydas.com](mailto:amlan@madebydas.com) *** # Skill 5: Performance Command Center *Close the gap between what happened and what to do next.* Skill5_Performance_Command_Diagram.png *** ***↑ Download performance-command-center.md file*** *** ## The Problem 79% of marketing leaders say current analytics tools provide insufficient actionability. 68% of ecommerce CMOs do not trust their marketing attribution data. 58% of leaders say key decisions are based on inaccurate or inconsistent data. And 51% of brands have discovered that channels they thought were profitable were actually losing money when measured against true contribution margin. The data exists. The dashboards exist. Triple Whale, Northbeam, Rockerbox, Google Analytics, Shopify Analytics, Meta Ads Manager, Klaviyo—every ecommerce brand has more data than it can process. The problem is not access. It is synthesis. It is turning six dashboards worth of numbers into three decisions about what to do this week. The typical weekly performance review at an ecommerce brand goes like this: someone pulls numbers from each platform, pastes them into a slide deck or sheet, presents in a meeting, and the team nods along. Decisions, if they happen at all, are made on gut feel informed by whichever number someone remembered from the deck. The deck goes into a shared drive. Nobody opens it again. This skill replaces that entire cycle. It takes raw performance data and produces narrative intelligence: not what happened, but what it means and what to do about it. ## The Workflow ```text theme={null} ┌──────────────────────────────────────────────────────────────┐ │ PERFORMANCE COMMAND CENTER │ │ │ │ INPUT: Weekly data from Shopify, Meta, Google, Klaviyo, │ │ Amazon, TikTok (CSV exports or pasted metrics) │ │ ───────────────────────────────────────────── │ │ │ │ Tier 1 ─── Executive Health Check │ │ (Revenue vs target, MER, biggest risk) │ │ │ │ │ Tier 2 ─── Channel Performance │ │ (Win / Loss / Learning per channel) │ │ │ │ │ Tier 3 ─── Creative Performance │ │ (Rank, fatigue signals, scale/kill/iterate) │ │ │ │ │ Tier 4 ─── Cross-Channel Synthesis │ │ (Attribution conflicts, budget reallocation) │ │ │ │ │ Tier 5 ─── Loop-Back Triggers │ │ (Which other skills should run and why) │ │ │ │ ───────────────────────────────────────────── │ │ OUTPUT: Weekly Action Brief with: │ │ ✓ Executive Summary (5 sentences max) │ │ ✓ Scorecard with trend arrows │ │ ✓ Channel Win/Loss/Learning analysis │ │ ✓ Prioritized Action List (task, owner, deadline, KPI) │ │ ✓ Loop-Back Triggers to other skills │ └──────────────────────────────────────────────────────────────┘ ``` * **Data Ingestion.** Accepts CSV exports or pasted metrics from Shopify, Meta Ads, Google Ads, Klaviyo, Amazon, and TikTok. Does not require API connections — works with whatever data format the operator can provide. * **Performance Narrative.** Translates raw metrics into plain-language intelligence. Instead of "CTR is 1.82%," the output says "CTR on static ads is up 14% week-over-week, driven by the new UGC-style creatives launched Tuesday. The video variants are underperforming — average watch time dropped below the 3-second threshold." * **Win/Loss/Learning Structure.** Each channel is analyzed through a three-part lens: what won (scale it), what lost (kill or fix it), and what was learned (test it next). This prevents the common failure mode of reviewing numbers without making decisions. * **Action List Generation.** Every performance review ends with a prioritized action list. Each item includes: the specific action, the single owner, the deadline, and the metric it impacts. No ambiguity, no shared ownership, no open-ended timelines. * **Loop-Back Triggers.** When performance data suggests a problem that another skill can address, the Command Center flags it: declining ad performance triggers a consumer intelligence refresh. Competitor gaining share triggers a competitive analysis. Creative fatigue triggers a new campaign brief. ## The SKILL.md ```markdown theme={null} --- name: performance-command-center description: > Synthesizes cross-channel ecommerce performance data into narrative intelligence and prioritized action lists. Turns dashboards into decisions. Use when asked to analyze performance data, review metrics, create weekly reports, or generate performance summaries. --- # Performance Command Center ## Context You are a senior growth strategist reviewing ecommerce performance data. Your job is not to summarize numbers — the operator can read numbers. Your job is to tell them what the data MEANS and what they should DO about it. Every insight must lead to an action. Read the Brand Brain for: KPI targets, channel priorities, and the metrics framework the team uses to make decisions. ## Input Handling Accept performance data in any format: - Shopify dashboard exports (revenue, orders, CVR, AOV, traffic) - Meta Ads exports (spend, CTR, CPM, CPA, ROAS by campaign/creative) - Google Ads exports (spend, CPC, ROAS, search term reports) - Klaviyo exports (sends, open rate, CTR, revenue per recipient) - Amazon reports (TACoS, ACoS, BSR, ad spend, organic vs paid split) - TikTok Ads exports (spend, CTR, CPA, creative performance) - Pasted summary metrics or described observations - Triple Whale / Northbeam / attribution tool exports Work with whatever is provided. Note gaps in data coverage. ## Analysis Framework ### Tier 1: Executive Health Check Answer these questions in 3-5 sentences: - Are we on track for revenue targets this period? - Is our MER (total revenue / total marketing spend) improving? - What is the single biggest risk to this period's performance? - [CUSTOMIZE: Your executive KPI targets] ### Tier 2: Channel Performance For EACH channel with data provided, analyze through the Win / Loss / Learning framework: WIN (scale it): - What is the single best performing element this period? - Why is it winning? (Be specific: which creative, which audience, which product, which offer) - Recommended scaling action with specific budget or volume increase LOSS (kill or fix it): - What is the single worst performing element this period? - Why is it losing? (Be specific: creative fatigue, audience exhaustion, competitive pressure, seasonality, technical issue) - Recommended action: kill, fix with specific change, or test LEARNING (test it next): - What is the most interesting pattern in this data? - What hypothesis does it suggest? - Recommended test: specific, measurable, time-bound ### Tier 3: Creative Performance If creative-level data is provided: - Rank creatives by efficiency (CPA or ROAS) - Identify creative fatigue signals (declining CTR over time) - Flag creatives approaching fatigue threshold (7-14 day window) - Recommend: scale, maintain, iterate, or retire for each ### Tier 4: Cross-Channel Synthesis Look across ALL channels for: - Revenue attribution conflicts between platforms - Channel interactions (email supporting paid performance or vice versa) - Budget reallocation opportunities (shift from low-MER to high-MER) - Emerging risks (over-dependence on single channel, rising CPMs) ### Tier 5: Loop-Back Triggers Based on the analysis, flag when other Growth Loop skills should run: - Consumer Intelligence Engine: When messaging resonance declines or a new customer objection pattern emerges - Trend Radar: When category-level shifts affect performance - Competitive Position Tracker: When a competitor appears to be gaining share or your positioning feels commoditized - Campaign Brief Architect: When creative fatigue hits threshold or a new campaign is needed based on performance gaps ## Output Format Structure as a WEEKLY ACTION BRIEF: 1. Executive Summary (5 sentences maximum) 2. Scorecard table: key metrics vs targets with trend arrows 3. Channel-by-channel Win/Loss/Learning analysis 4. Creative performance ranking (if data available) 5. Budget reallocation recommendations (if warranted) 6. Prioritized Action List: | Priority | Action | Owner | Deadline | KPI Impact | 7. Loop-Back Triggers (which other skills to run and why) ## Guardrails - Never present data without interpretation. - Never interpret data without a recommended action. - Every action item must have ONE owner (not 'the team'). - Flag attribution uncertainty. Use language like 'platform-reported' vs 'blended/verified' when appropriate. - Note when data is insufficient for confident recommendations. - [CUSTOMIZE: Your team's decision-making framework and roles] ``` ## See It Work: Sample Output Here is a condensed Weekly Action Brief for a hypothetical DTC wellness brand: > **EXECUTIVE SUMMARY:** Revenue hit \$187K this week, 8% above target, driven primarily by the new UGC static creatives launched on Meta Tuesday. MER improved to 4.2x (target: 3.5x). The primary risk is Klaviyo deliverability — open rates dropped 22% week-over-week, suggesting a list hygiene issue. Immediate action required on email. The magnesium launch campaign brief (Skill 4) should be prioritized given the strong early signal on Meta. > > **META ADS — Win / Loss / Learning:** > > Win: UGC static set B ("2pm fog" angle) delivered \$18.40 CPA vs. \$28 target. Scale: increase daily budget 30% and duplicate to Advantage+ campaign. > > Loss: Video variants averaged 1.8 seconds watch time (below 3-second threshold). The polished production style is not connecting. Kill: pause all 3 video creatives and reallocate budget to static UGC. > > Learning: The "cabinet replacement" angle (Set C) has a CTR 40% above average but CPA 15% above target. Hypothesis: strong hook but weak landing page match. Test: create a dedicated LP for this angle. > > **PRIORITIZED ACTION LIST:** > > ``` > ┌──────┬──────────────────────────────────────────┬────────┬───────────┬─────────────────┐ > │ Pri │ Action │ Owner │ Deadline │ KPI Impact │ > ├──────┼──────────────────────────────────────────┼────────┼───────────┼─────────────────┤ > │ P1 │ Run Klaviyo list cleaning + re-engage │ Sarah │ Friday │ Email rev share │ > │ P2 │ Scale Meta UGC Set B budget +30% │ Mike │ Tomorrow │ CPA efficiency │ > │ P3 │ Pause 3 video creatives, realloc to stat │ Mike │ Tomorrow │ Spend efficiency│ > │ P4 │ Brief new LP for "cabinet replacement" │Creative│ Next Mon │ Meta CVR │ > └──────┴──────────────────────────────────────────┴────────┴───────────┴─────────────────┘ > ``` > > **LOOP-BACK TRIGGERS:** > → Campaign Brief Architect: Brief 3 new creative concepts to replace paused video variants. Use Consumer Intelligence Engine output for messaging angles. > → Competitive Position Tracker: GlowBase appears to be increasing Meta spend (higher impression share on shared keywords). Run competitive scan this week. ## Customization Notes * **Data format flexibility:** The skill works with whatever you can provide. A full CSV export from every platform is ideal but not required. Pasting key metrics into the conversation and describing trends works for quick weekly reviews. The quality of recommendations scales with data completeness, but even partial data produces actionable output. * **Metric customization:** Every brand has its own metric hierarchy. Adjust the KPI targets in the \[CUSTOMIZE] sections to match your specific benchmarks. A brand targeting 5x MER will get very different recommendations than one targeting 3x. A brand where email drives 40% of revenue will see Klaviyo issues flagged differently than one where email is 15%. * **Frequency:** Run weekly for ongoing performance management. Run daily during major campaigns, launches, or promotional periods. Run monthly for deeper cross-channel synthesis and budget reallocation analysis. > **Want this built for your brand?** DAS builds Performance Command Centers with automated data ingestion from Shopify, Meta, Klaviyo, and Amazon, scheduled weekly synthesis to your inbox, and Slack alerts for metric anomalies. → [amlan@madebydas.com](mailto:amlan@madebydas.com) *** # The Loop in Action: One Week, One Full Cycle Here is how the five skills connect in a single cycle—the exact workflow a brand can run in a single week. Loop_In_Action_Timeline.png **Monday: Performance Command Center (Skill 5) runs the weekly review.** The team pastes last week's data from Shopify, Meta, and Klaviyo. The skill identifies that conversion rate dropped 18% on the top-performing product page. Ad creative CTR is declining across all sets — creative fatigue is setting in. The action brief flags two loop-back triggers: run Consumer Intelligence Engine to check for emerging objections, and brief new creative through Campaign Brief Architect. **Tuesday: Consumer Intelligence Engine (Skill 1) analyzes the latest 90 days of reviews.** The team exports reviews from Shopify and Amazon. The skill surfaces a new pattern: customers in the last 30 days are increasingly mentioning "aftertaste" as a negative — a theme that did not appear in the previous quarter's analysis. The messaging matrix reveals a shift: the strongest purchase trigger has moved from "ingredient quality" to "peer recommendation." The product team gets the feedback priority list. The content team gets updated ad copy angles. **Wednesday: Trend Radar (Skill 2) validates a signal the team noticed.** A team member saw magnesium drinks gaining traction on TikTok. They feed TikTok data, Google Trends exports, and Amazon search term reports into the Trend Radar. The skill validates the signal across three platforms, scores it at 100/125 on the opportunity matrix, and classifies it as a macrotrend. The recommendation: brief a magnesium SKU extension within 30 days and create educational content this week. The team also gets a contrarian call: mushroom coffee, which they were considering, is showing declining velocity. **Thursday: Competitive Position Tracker (Skill 3) scans the landscape.** The team feeds in Meta Ad Library screenshots and competitor pricing data. The skill reveals that the primary competitor has shifted messaging from clinical efficacy to community-driven social proof — a significant repositioning. The whitespace analysis shows that no competitor is running ingredient comparison content, despite it being one of the strongest hooks identified in the Consumer Intelligence Engine output. The battlecard updates automatically. **Friday: Campaign Brief Architect (Skill 4) generates the next campaign brief.** It pulls together everything from the week: the new peer recommendation messaging angle from Skill 1, the magnesium trend validation from Skill 2, the ingredient comparison whitespace from Skill 3, and the creative fatigue data from Skill 5. The output is a complete brief with channel-specific deliverables for Meta (3 new static concepts, 1 comparison carousel), Email (2 education-focused sends), and Amazon A+ (updated modules for the hero SKU). The creative team starts production Monday. > **This Loop is one cycle, and it only took a week.** Without the skills, this same cycle takes double or triple the time, and requires a dedicated resources (analyst, strategist, account managers, etc.) to do the work. Over a year, the skills-enabled brand runs 52 of these cycles. The manual brand runs 8–12. **The compounding intelligence gap becomes the competitive moat.** > *** # What's Next Get_It_Built_DAS_CTA_1.png ### Path 1: Install and Run Download the five skills from the Quick Start section at the top. Upload to Claude. Build your Brand Brain from the template. Start running the loop. You have everything you need. ### Path 2: Get It Built You want the full Growth Loop built for your brand. Customized Brand Brain with competitive research and persona development. Tailored skills connected to your specific data sources. Integration with your team's workflow and tools. Ongoing optimization as your brand scales. That is what DAS is here for: functioning as the orchestration layer of building the systems your growth team can run on. > **Ready to talk?** > > Reach out directly: [amlan@madebydas.com](mailto:amlan@madebydas.com) > > Or connect with [Amlan on LinkedIn](https://www.linkedin.com/in/amlankdas/) and comment GROWTHLOOP on the post where you found this playbook. *** # The Nano Banana 2 Ad Prompt Playbook Source: https://intelligence.madebydas.com/playbooks/nano-banana-ad-prompts 25 copy-paste prompts for production-ready brand assets using NanoBanana 2 — the prompt architecture, quality gates, and batch workflows for DTC creative at scale. **25 Copy-Paste Prompts for Production-Ready Brand Assets Using NanoBanana 2** *** # **Before you dive in…** This playbook is dense by design. It is meant to be returned to, not skimmed through one time. Bookmark it, build with it, come back when you are stuck. If you want a custom workflow built for your brand, a specific prompt architecture developed, or just a second set of eyes on what you are generating — reach out. **Amlan Das, Founder — DAS Audience Development |** [amlan@madebydas.com](mailto:amlan@madebydas.com) *** # Part 1: Why This Playbook Exists This playbook is not for brands still figuring out who their customer is. It is for brands that already have a product people want, a customer who buys it, and a creative bottleneck that is quietly killing their growth. If you are spending more time producing ads than analyzing which ones work, this is for you. **A line worth drawing before we start:** AI does not replace your brand. Your homepage hero, your packaging, your campaign flagship image, those all still require a human photographer with full control and full accountability. This playbook is not touching those. What this playbook solves is everything else. The 50 Meta variants you need to find efficiency. The email headers. The seasonal swaps. The carousel cards. The Amazon PDP badges. The UGC-style scroll-stoppers. The assets that feed the algorithm and fund the shoots that actually matter. **The math that broke the old model.** A traditional commercial shoot costs \$2K–\$10K per day and yields roughly 20 usable assets — about \$250 each. To give Meta and TikTok the 100 variants they need to find efficiency, you are spending \$25K in production before a single impression is bought. For most DTC brands, that CAC load is unsustainable before you account for media spend, fulfillment, or the cost of being wrong. NanoBanana 2 (Gemini 3.1 Flash Image, released Feb 26, 2026) changed the production economics. It solved the gibberish problem, prior models could generate beautiful images but failed at legible packaging text, which made them useless for conversion ads. NanoBanana 2 offers native text rendering, 4K output, and near-professional lighting control at 2–5x the speed of human workflows. **But volume without direction is just noise.** The brands wasting money on AI creative are treating it like a vending machine; you feed it a generic prompt, run the output, wonder why it doesn't convert. The brands extracting real value are using it the way a creative director uses a brief: with specificity about the product, the placement, and the person seeing it. Before you run a single prompt in this playbook, know which customer you are talking to. Your best customers are not the same as your most recent customers. A lifestyle shot that resonates with your VIP buyer will not convert a first-time visitor. The prompts here are flexible enough to serve both — but only if you know the difference before you generate. That distinction is the whole game. # Part 2: The Brand Research Prompt Before you run a single prompt from this playbook, you need to teach the AI your brand. Create a Claude Project for your brand and run this prompt. It will research everything NanoBanana 2 needs to match your visual identity perfectly. This is a one-time setup that pays dividends across every prompt in this playbook. ```text theme={null} INPUT VARIABLES: Target Brand Name: [INSERT BRAND NAME HERE] Target URL: [INSERT URL HERE] Role: Act as a Senior Brand Strategist conducting a full reverse-engineering of the target brand's visual and verbal identity. Objective: Create a comprehensive Brand DNA document that will be used to write highly specific AI image generation prompts. Every detail matters because the output will be fed into an image model that needs exact specifications. PHASE 1: EXTERNAL RESEARCH Use web search to find the source of truth for this brand: 1. Design credits: Search for "who designed [Brand] branding", "[Brand] design agency case study", "[Brand] rebrand" 2. Public brand assets: Search for "[Brand] brand guidelines pdf", "[Brand] press kit", "[Brand] media kit", "[Brand] style guide" 3. Typography: Search for "[Brand] font", "[Brand] typeface", "what font does [Brand] use" 4. Colors: Search for "[Brand] brand colors", "[Brand] hex codes", "[Brand] color palette" 5. Packaging: Search for "[Brand] packaging design", "[Brand] unboxing", "[Brand] product photography" 6. Advertising: Search "[Brand]" in Meta Ad Library (facebook.com/ads/library) for current ad creative styles 7. Press and positioning: Search for "[Brand] brand story", "[Brand] founding story", "[Brand] mission" PHASE 2: ON-SITE ANALYSIS Visit the Target URL and analyze: 1. Voice and Tone: Read hero copy, About page, and product descriptions. Give me 5 distinct adjectives. 2. Photography Style: Describe lighting, color grading, composition, and subject matter. 3. Typography on site: Headline weight, body weight, letter-spacing, distinctive treatments. 4. Color application: Primary vs accent usage. Background colors. CTA color. 5. Layout density: Airy or dense? Grid-based or organic? 6. Packaging details: Physical appearance (materials, colors, shape, label placement, textures, translucency, matte vs gloss). PHASE 3: COMPETITIVE CONTEXT Search for 2-3 direct competitors and note visual differentiation. PHASE 4: OUTPUT Combine into this format: BRAND DNA DOCUMENT ================== BRAND OVERVIEW - Name / Tagline / Design Agency / Voice Adjectives [5] / Positioning / Competitive Differentiation VISUAL SYSTEM - Primary Font / Secondary Font / Primary Color [hex] / Secondary Color [hex] / Accent Color [hex] / Background Colors / CTA Color and Style PHOTOGRAPHY DIRECTION - Lighting / Color Grading / Composition / Subject Matter / Props and Surfaces / Mood PRODUCT DETAILS - Physical Description / Label-Logo Placement / Distinctive Features / Packaging System AD CREATIVE STYLE - Typical formats / Text overlay style / Photo vs illustration / UGC usage / Offer presentation IMAGE GENERATION PROMPT MODIFIER Write a single 50-75 word paragraph to prepend to any image prompt to match this brand's visual identity. Include exact colors, font descriptions, photography direction, and mood. ``` The Brand DNA Document output becomes your master reference. When using any prompt in this playbook, prepend the IMAGE GENERATION PROMPT MODIFIER to force brand consistency across every asset. # Part 3: The Prompt Architecture We do not rely on "vibes" for prompting; we use a rigid 5-parameter architecture. This ensures consistency whether we are generating assets for a luxury handbag or a consumer electronic device. **1. Subject Definition:** The core object. Be hyper-specific about materials and textures. *Example:* "A matte-black iPhone 16Pro phone case with a subtle lip-design around the camera lens to protect it" or "A heavyweight cotton hoodie in oatmeal heather with ribbed cuffs." **2. Environment/Lighting:** Where the product lives and how light hits it. *Example:* "Golden hour sunlight streaming through sheer curtains," "Hard flash studio lighting on seamless pink background," or "Dappled forest light on mossy stone." **3. Stylization:** The camera lens and artistic direction. *Example:* "Shot on 35mm film, grainy texture, f/1.8 aperture for shallow depth of field" or "Hyper-realistic 4K product render, sharp focus, macro lens." **4. Text/Graphics Layer:** The copy on the product or overlay. *Example:* "Text 'SUMMER SALE' in bold sans-serif white font" or "Label reads 'ORGANIC' in gold foil stamping." **5. Technical Constraints:** Aspect ratios and negative space requirements. *Example:* "--ar 4:5 --no blur --leave upper 30% empty for copy." **6. The Sixth Parameter: Who Is Seeing This** Before you write a single prompt, answer one question: which customer is this for? Not your average customer. A specific one. Your VIP reorder buyer responds to different visual cues than someone seeing your brand for the first time on TikTok. Your retail customer is not your DTC customer. These people exist in the same database but they do not live in the same world. The prompts in this playbook are flexible. That flexibility is only useful if you constrain it intentionally. "Lifestyle shot" means something different depending on whether your buyer is a 34-year-old CrossFit coach or a 58-year-old interior designer. The prompt architecture handles the technical execution. You handle the targeting logic. If you are generating assets without a specific segment in mind, you’ll be gambling, not iterating. **The Quality Checklist** Before exporting an asset for ad testing, run it against this checklist. If it fails any point, regenerate. * **Text Spelling:** Verify the brand name and key claims (ex: "50g Protein," "100% Cotton") are spelled correctly. * **Product Geometry:** AI struggles with symmetry. Check handle loops on bags, zippers on jackets, and the circularity of rims/lids. * **Label/Logo Legibility:** Ensure text wraps correctly around the curvature of the product surface; it should not look like a flat PNG pasted on top. * **Human Elements:** If hands or models are present, check anatomy. Look for 5 fingers, natural joint articulation, and realistic skin texture. **A Note on Variables** All 25 prompts in this playbook use \[BRACKETED VARIABLES]. You must swap these with your specific product details. While these prompts are optimized for NanoBanana 2, the architecture functions on any frontier image model (Midjourney v7, OpenAI Sora Stills, etc.). *** # Part 4: Case Study — Barebells Protein Soda To show you how this playbook works in practice, we ran five prompts against a real product: Barebells Protein Soda (Sweet Cherry). Barebells is a Swedish functional food brand known primarily for their protein bars. Their Protein Soda line is a newer product; a carbonated, sugar-free drink with 10g of protein and 200mg of caffeine in a slim 12oz can. The packaging features a light blue can with retro pink script lettering and a purple accent stripe. We chose Barebells because it is a strong test case: the can has complex text rendering (script font, multiple text elements, attribute callouts), a distinctive color palette, and a physical form factor (slim aluminum can with condensation) that challenges AI image models. We selected five prompts from this playbook that cover the most common DTC creative needs: a studio hero, an ingredient burst, a lifestyle/UGC shot, a sensory macro, and an Amazon-compliant PDP image. Here is what we generated. > **ⓘ A note about taste:** Your first outputs may not feel ‘on-brand’ and that's expected. AI image generation is a directed craft, not a vending machine. The prompts in this playbook are starting points. Lighting, mood, composition, color grading, all of it responds to specificity. If something looks off, the answer is iteration, not abandonment. The brands getting the most out of these tools treat prompting like a creative brief: the more precise the direction, the tighter the output. **Prompt 1 — Product Hero Shot** This is the foundation of any product page. We needed a clean, studio-quality isolated shot that could serve as the primary image on a Shopify PDP or retargeting ad. ```text theme={null} /imagine prompt: Subject: A slim 12oz light blue aluminum can of Barebells Protein Soda with pink retro script lettering reading "Protein Soda", purple accent stripe, and "Sweet Cherry" in magenta text. Condensation droplets visible on the can surface. Label Text: The brand name "Barebells" is clearly visible in pink script at the top. Large bold text reads "PROTEIN 10g". Lighting: High-key studio lighting, soft shadows, 50mm lens, sharp focus. Background: A smooth soft pink gradient background. Elements: Floating 3D badge to the right of the product reading "SUGAR FREE". Style: Commercial product photography, hyper-realistic, 8k resolution, ray-traced reflections. --ar 4:5 --v 6.0 --style raw ``` barbell - Product Hero Shot.png The model rendered the brand name in script, placed the "Sugar Free" badge correctly, and maintained the light blue and pink color palette. The soft pink gradient background complements the can without competing with it. **Prompt 3 — Ingredient Explosion** This is the "scroll-stopper" for Meta carousel ads. We wanted cherries, liquid splash, and carbonation bursting around the can to visually validate the "Sweet Cherry" flavor claim. ```text theme={null} Professional commercial photography of Barebells Protein Soda slim light blue aluminum can with pink retro script lettering bursting through a cloud of fresh whole cherries, cherry halves with visible pits, frozen carbonation bubbles, and ice crystal shards. High-speed photography style, 1/8000s shutter speed, frozen motion droplets of sparkling liquid suspended in air. Macro lens 100mm, f/11 for sharp focus on the label. Lighting: High-key studio lighting with rim light to accentuate transparency in liquid and carbonation. Background: Solid color #F0E6EF matching a soft pink-lavender palette but slightly desaturated. 8k resolution, hyper-realistic textures. --ar 4:5 --stylize 250 ``` barbell - Ingredient Explosion.png The frozen-motion cherries and liquid splashes create immediate visual energy. This type of image outperforms static product shots in Meta carousel positions 2-3 because it communicates flavor and freshness without the viewer needing to read a single word. **Prompt 5 — Hard Flash Aesthetic** For TikTok and Instagram, polished studio shots get scrolled past. This prompt generates a raw, "found in the wild" image that looks like someone snapped it at the gym between sets. ```text theme={null} Amateur flash photography of Barebells Protein Soda light blue can with pink script lettering sitting on a messy gym bench. Surrounding objects: a pair of lifting gloves, a phone with earbuds tangled around it, a half-used roll of athletic tape, and a shaker bottle cap. Lighting: Direct harsh camera flash, hard shadows directly behind the object. Aesthetic: Disposable camera, Fujifilm QuickSnap style, high contrast, slightly overexposed highlights. Angle: High angle looking down, slightly dutch angle/tilted. Raw, unedited, candid vibe. --ar 9:16 --style raw --s 50 ``` barbell - Hard Flash Aesthetic.png This is the most valuable creative in the batch for paid social. The gym context, harsh flash, and messy surroundings signal "real person, real life" to the algorithm and the viewer. It does not look like an ad, which is exactly why it works. **Prompt 6 — Texture Macro** This extreme close-up sells the sensory experience of a cold can. Use it as a detail card in carousels, a text overlay background, or a story frame. ```text theme={null} Extreme macro close-up photography of condensation on a cold light blue aluminum can. Subject: Dense water droplets beading on the curved metallic surface, with a sliver of pink retro script lettering visible through the moisture. Camera: 100mm macro lens, f/2.8 depth of field. Lighting: Caustics and refraction lighting to highlight translucency and moisture. Details: Visible irregular droplet sizes, streaks where condensation has run, sharp focus on the largest droplet, soft bokeh background showing blurred pink and blue. Sensory details: ice-cold, refreshing, thirst-quenching. 8k resolution, Unreal Engine 5 render style. --ar 1:1 --tile ``` barbell - Texture Macro.png You can almost feel the cold aluminum. This type of asset compensates for the sensory deficit of online shopping — the viewer cannot hold the product, so you bring the texture to them. **Prompt 8 — Amazon PDP Hero** Amazon requires a pure white background for main listing images. This prompt adds vector attribute badges to maximize the information density within the frame, which directly impacts mobile CTR. ```text theme={null} /imagine prompt: A slim 12oz light blue aluminum can of Barebells Protein Soda with pink retro script lettering, purple accent stripe, condensation droplets on surface, "Sweet Cherry" flavor text in magenta, shot on pure white #FFFFFF background, studio lighting, soft shadows directly underneath, high key photography, 8k, sharp focus. To the right of the product, three flat vector icons arranged vertically: 1) a circle with text "10g Protein", 2) a circle with text "Sugar Free", 3) a circle with text "200mg Caffeine". Icons are minimal, flat design, clean lines --ar 1:1 --stylize 50 --v 6.0 --no background noise, grey cast, vignetting ``` barbell - Amazon PDP Hero.png The model placed the attribute badges correctly and maintained the pure white background required by Amazon. The three key selling points — protein content, sugar-free, and caffeine — are immediately visible without requiring the shopper to read the listing. **Takeaway** Five prompts. One product. Under 30 minutes of total generation time. The cost was negligible. These five assets cover the core creative needs for a product launch: a hero image, a scroll-stopper, a UGC-style social ad, a sensory detail shot, and a marketplace listing. Scale this across your full product line and you have a complete creative library without a single studio booking. *** # Part 5: The 25 Prompts **Prompt 1 — The Product Hero Shot** Studio-quality isolated product on a clean or gradient background. Drives highest CTR on Amazon and Google Shopping for primary PDP images or retargeting ads. **Input Variables** * \[Product Description]: \[ex: A matte black protein powder tub with gold foil lettering] * \[Brand Name]: \[Your Brand Name] * \[Benefit Text]: \[ex: 25G PROTEIN] * \[Color]: \[Background color, ex: soft blue] * \[Badge Text]: \[ex: KETO] **The Prompt** ```text theme={null} /imagine prompt: Subject: [Product Description] centered in frame. Label Text: The brand name "[Brand Name]" is clearly visible. Large bold text reads "[Benefit Text]". Lighting: High-key studio lighting, soft shadows, 50mm lens, sharp focus. Background: A smooth [Color] gradient background. Elements: Floating 3D badge to the right of the product reading "[Badge Text]". Style: Commercial product photography, hyper-realistic, 8k resolution, ray-traced reflections. --ar 4:5 --v 6.0 --style raw ``` **Prompt 2 — The Lifestyle Context Shot** Product in an aspirational setting selling the identity, not just the product. Ideal for Instagram Stories and TikTok ads where organic-looking content outperforms polished studio work. **Input Variables** * \[Target Demographic]: \[ex: fit woman in her 30s] * \[Product]: \[ex: green juice bottle] * \[Action]: \[ex: laughing while walking] * \[Brand Name]: \[Your Brand Name] * \[Setting]: \[ex: sunny outdoor farmers market] **The Prompt** ```text theme={null} /imagine prompt: Subject: A [Target Demographic] holding a [Product] casually. Focus is on the product. Action: She is [Action]. Label Text: The text "[Brand Name]" is legible on the bottle. Setting: A [Setting], depth of field, blurred background. Lighting: Natural sunlight, lens flare, golden hour. Style: UGC style, iPhone photography, candid, authentic, slightly grainy. --no messy background, deformed hands, extra fingers --ar 9:16 ``` **Prompt 3 — The Ingredient Explosion** High-energy shot of product bursting through its raw ingredients. Use for carousel cards 2-3 on Meta to visually validate claims made in the hook. **Input Variables** * \[Product Name]: \[Your Product Name] * \[Product Packaging Description]: \[ex: amber glass bottle with white label] * \[Key Ingredients]: \[ex: sliced lemons, raw ginger root, turmeric powder] * \[Liquid Type]: \[ex: water, oil, serum] * \[Hex Code]: \[Background color code, ex: #F4F4F4] **The Prompt** ```text theme={null} Professional commercial photography of [Product Name] [Product Packaging Description] bursting through a cloud of [Key Ingredients]. High-speed photography style, 1/8000s shutter speed, frozen motion droplets of [Liquid Type] suspended in air. Macro lens 100mm, f/11 for sharp focus on the label. Lighting: High-key studio lighting with rim light to accentuate transparency in liquid. Background: Solid color [Hex Code] matching the brand palette but slightly desaturated. 8k resolution, hyper-realistic textures. --ar 4:5 --stylize 250 ``` **Prompt 4 — The Social Ad Creative (Meta)** Performance ad with headline and CTA text baked into the image for native lighting integration. Bypasses Canva/Figma workflows. **Input Variables** * \[Product Name]: \[Your Product Name] * \[Product]: \[Visual description of product] * \[Texture]: \[Surface material, ex: marble, wooden, concrete] * \[Color]: \[Wall color] * \[Headline]: \[ex: DEEP SLEEP] * \[Text Color]: \[Color of headline text] * \[CTA]: \[ex: SHOP NOW] **The Prompt** ```text theme={null} A vertical social media ad creative for [Product Name]. Center composition: [Product] sitting on a [Texture] surface. Background: Minimalist [Color] wall. Text Integration: Large, bold, sans-serif text floating behind the product reading "[Headline]" in [Text Color]. Bottom overlay text: A button shape reading "[CTA]" in contrasting color. Lighting: Soft window light from the right, casting realistic shadows from the product onto the text. Photorealistic, 4k, sharp text rendering. --ar 4:5 --v 6.0 ``` **Prompt 5 — The Hard Flash Aesthetic** Gritty, disposable-camera-style photo that signals peer content rather than brand content. Outperforms polished studio shots for younger audiences on TikTok and Instagram. **Input Variables** * \[Product Name]: \[Your Product Name] * \[Surface]: \[ex: bedside table, bathroom counter] * \[Context Items]: \[ex: car keys, half-drunk glass of water, open magazine, hair ties] **The Prompt** ```text theme={null} Amateur flash photography of [Product Name] sitting on a messy [Surface]. Surrounding objects: [Context Items]. Lighting: Direct harsh camera flash, hard shadows directly behind the object. Aesthetic: Disposable camera, Fujifilm QuickSnap style, high contrast, slightly overexposed highlights. Angle: High angle looking down, slightly dutch angle/tilted. Raw, unedited, candid vibe. --ar 9:16 --style raw --s 50 ``` **Prompt 6 — The Texture Macro** Extreme close-up of product texture to compensate for the sensory deficit of online shopping. Use as detail cards in carousels or text overlay backgrounds. **Input Variables** * \[Texture Description]: \[ex: thick white cream, carbonated liquid] * \[Specific Detail]: \[ex: a smear of cream, condensation droplets on a cold can] **The Prompt** ```text theme={null} Extreme macro close-up photography of [Texture Description]. Subject: [Specific Detail]. Camera: 100mm macro lens, f/2.8 depth of field. Lighting: Caustics and refraction lighting to highlight translucency and moisture. Details: Visible pores in the cream, irregular bubble sizes, sharp focus on the texture peak, soft bokeh background. Sensory details: rich, luscious, hydrating, thirst-quenching. 8k resolution, Unreal Engine 5 render style. --ar 1:1 --tile ``` **Prompt 7 — The Seasonal Campaign Variant** Same product placed in a seasonal context to bypass physical production. Bakes seasonal copy directly into the image. **Input Variables** * \[Product Description]: \[Visual description of product] * \[Seasonal Elements]: \[ex: dried maple leaves, miniature pumpkins, cinnamon sticks] * \[Seasonal Lighting Condition]: \[ex: warm golden hour sunlight casting long shadows] * \[Seasonal Copy]: \[ex: FALL FAVORITES] **The Prompt** ```text theme={null} /imagine prompt: professional commercial photography of [Product Description] centered on a podium, surrounded by [Seasonal Elements], [Seasonal Lighting Condition], depth of field, 8k resolution, photorealistic texture --ar 4:5 --stylize 250 --v 6.0 --text "[Seasonal Copy]" written in bold serif font floating behind the product ``` **Prompt 8 — The Amazon PDP Hero** Amazon-compliant main image on a pure white background with vector attribute badges. Maximizes mobile CTR with high frame fill. **Input Variables** * \[Product Description]: \[Visual description of product] * \[Attribute 1]: \[ex: 20g Protein] * \[Attribute 2]: \[ex: Keto] * \[Attribute 3]: \[ex: 0g Sugar] **The Prompt** ```text theme={null} /imagine prompt: [Product Description] shot on pure white #FFFFFF background, studio lighting, soft shadows directly underneath, high key photography, 8k, sharp focus. To the right of the product, three flat vector icons arranged vertically: 1) a circle with text "[Attribute 1]", 2) a circle with text "[Attribute 2]", 3) a circle with text "[Attribute 3]". Icons are minimal, flat design, clean lines --ar 1:1 --stylize 50 --v 6.0 --no background noise, grey cast, vignetting ``` **Prompt 9 — The Email Header/Banner** Wide panoramic asset for email headers (Klaviyo, Mailchimp) with intentional negative space for HTML text overlay. Minimizes vertical pixel usage. **Input Variables** * \[Product Description]: \[Visual description of product] * \[Surface Material]: \[ex: white marble counter] * \[Background Texture]: \[ex: soft blurred kitchen tiles] **The Prompt** ```text theme={null} /imagine prompt: panoramic wide shot of [Product Description] placed on the far left third of the frame, sitting on a [Surface Material]. The right two-thirds of the image is empty negative space with [Background Texture]. Soft natural lighting coming from the left, minimalist composition, commercial aesthetic --ar 3:1 --stylize 100 --v 6.0 --no text, clutter in negative space ``` **Prompt 10 — The In-Situ Context Shot** "Lived-in" realism that builds trust through intentional clutter and smartphone aesthetics. Ideal for whitelist ads or review-style creatives. **Input Variables** * \[Product Description]: \[Visual description of product] * \[Messy Environment]: \[ex: cluttered bathroom shelf with other toiletries] **The Prompt** ```text theme={null} /imagine prompt: amateur iPhone photo of [Product Description] sitting on a [Messy Environment]. harsh flash photography, slight motion blur, unposed, messy, authentic, user generated content style, realistic texture, fingerprints on bottle --ar 9:16 --stylize 50 --v 6.0 --no professional lighting, bokeh, studio setup ``` *** **Prompt 11 — The Flat Lay Arrangement** Perfect for bundles, starter kits, and "what's inside" ads. This overhead angle organizes chaos into a satisfying, logical structure that performs exceptionally well for skincare routines and subscription boxes. **Input Variables** * \[Product List]: \[ex: Serum bottle, moisturizer jar, face wash tube] * \[Props]: \[ex: Eucalyptus leaves, raw ingredients, textured linen] * \[Background]: \[ex: Marble slab, pastel pink matte surface] **The Prompt** ```text theme={null} Overhead knolling photography of [PRODUCT LIST] arranged neatly on a [BACKGROUND]. Symmetrical composition, organized layout. Surrounded by scattered [PROPS] for texture. Soft, shadowless studio lighting, 90-degree top-down angle. High resolution, f/16 aperture for full sharpness, commercial product photography, 8k, highly detailed textures. ``` *** **Prompt 12 — The Unboxing Shot** Captures the dopamine hit of a package arrival. Use this for middle-of-funnel retargeting or welcome flows to simulate the customer experience and build anticipation. **Input Variables** * \[Packaging Type]: \[ex: Branded cardboard mailer, rigid luxury box] * \[Inner Material]: \[ex: Crinkled tissue paper, custom inserts] * \[Product]: \[The main item being revealed] **The Prompt** ```text theme={null} POV photography of a [PACKAGING TYPE] being opened on a clean wooden table. Inside reveals [PRODUCT] nestled in [INNER MATERIAL]. "Just opened" aesthetic, messy but aesthetic tissue paper. Natural window light coming from the side, soft shadows. Focus on the product logo, shallow depth of field (f/2.8) to blur the table edges. High-end DTC unboxing experience, photorealistic, 4k. ``` *** **Prompt 13 — The Color/Variant Grid** Essential for collection launches. This shot proves range and variety, helping customers self-select their preference immediately. High conversion potential for fashion basics and cosmetics. **Input Variables** * \[Product Type]: \[ex: T-shirts, lipstick tubes, soda cans] * \[Color Palette]: \[ex: Earth tones, neon variety, pastel gradient] * \[Layout Style]: \[ex: 3x3 grid, diagonal repeating pattern] **The Prompt** ```text theme={null} Commercial product photography of [PRODUCT TYPE] in a variety of [COLOR PALETTE]. Arranged in a perfect [LAYOUT STYLE]. Uniform lighting, consistent shadows, clean solid background color that contrasts with the products. Repetitive composition, pop-art influence. Shot with a 50mm lens, sharp focus edge-to-edge, vibrant colors, 8k resolution. ``` *** **Prompt 14 — The Size/Scale Reference** Overcomes the "digital barrier" by showing physical context. Use this for jewelry, tech accessories, or travel goods where size ambiguity kills conversion rates. **Input Variables** * \[Product]: \[Your product] * \[Reference Object]: \[ex: iPhone 15, coffee mug, human hand, quarter] * \[Setting]: \[ex: Desktop, bedside table, cafe table] **The Prompt** ```text theme={null} Close-up product shot of [PRODUCT] placed directly next to a [REFERENCE OBJECT] for scale context. Resting on a [SETTING] surface. Realistic sizing, everyday lifestyle context. Natural daylight, neutral colors. 35mm lens, sharp focus on the product, slight bokeh on the background. Photorealistic, believable composition, high detail. ``` *** **Prompt 15 — The Model Wearing/Using** The "Social Proof" visual. Shows the product in action to help the viewer visualize themselves owning it. Critical for apparel, wearables, and outdoor gear. **Input Variables** * \[Model Description]: \[ex: Woman in her 30s, athletic build] * \[Action]: \[ex: Jogging, typing on laptop, drinking coffee] * \[Product Visibility]: \[ex: Wearing smart glasses, holding a tumbler] * \[Setting]: \[ex: Urban rooftop, coffee shop, hiking trail] **The Prompt** ```text theme={null} Lifestyle photography of a [MODEL DESCRIPTION] [ACTION] while using [PRODUCT VISIBILITY]. Candid moment, genuine expression, not posing for the camera. [SETTING] background, golden hour lighting with lens flare. Shot on 85mm portrait lens, f/1.8 aperture, subject in sharp focus, creamy background bokeh. Cinematic color grading, high-end fashion editorial style. ``` *** **Prompt 16 — The Product Lineup** The "Family Portrait" of your brand. Builds authority by showing catalog depth. Use this for top-of-funnel brand awareness ads to show you are a serious player, not a one-product dropshipper. **Input Variables** * \[Product Range]: \[ex: Full skincare line: cleanser, toner, moisturizer] * \[Podium/Surface]: \[ex: White stone steps, wooden risers, glass reflection] * \[Vibe]: \[ex: Clinical and clean, warm and organic] **The Prompt** ```text theme={null} Eye-level group shot of [PRODUCT RANGE] arranged on [PODIUM/SURFACE]. Hero product in the center, secondary products tiered behind. [VIBE] aesthetic. Studio lighting with rim light to separate products from background. High fidelity, commercial advertising standard, crystal clear text rendering, 8k resolution. ``` *** **Prompt 17 — The Comparison Shot** Visualizes your value proposition. While you can't use real competitor logos in AI, you can use "generic" equivalents to highlight your premium materials or superior design. **Input Variables** * \[Your Product]: \[Specific visual description of your premium item] * \[Generic Product]: \[Description of cheap/standard alternative] * \[Contrast Feature]: \[ex: Glowing vs. dull, sturdy vs. flimsy] **The Prompt** ```text theme={null} Side-by-side comparison photography. Left side: A generic, dull [GENERIC PRODUCT] with flat lighting. Right side: A premium, vibrant [YOUR PRODUCT] with [CONTRAST FEATURE] and luxury lighting. Visualizing the difference in quality. Split composition, high contrast. 4k, hyper-realistic, advertising style comparison. ``` *** **Prompt 18 — The Gift-Ready Shot** Capitalizes on Q4 and holiday urgency. This prompt styles your product not as a utility, but as a present. Essential for "Gift Guide" landing pages and November/December ad spend. **Input Variables** * \[Product]: \[Your product] * \[Wrapping Elements]: \[ex: Satin ribbon, gift tag, bow] * \[Holiday Context]: \[ex: Christmas tree lights in bokeh, Valentine's confetti] **The Prompt** ```text theme={null} Luxury product photography of [PRODUCT] styled for gifting. Wrapped with [WRAPPING ELEMENTS]. Festive atmosphere with [HOLIDAY CONTEXT] in the blurred background. Warm, cozy lighting, sparkling reflections. Shot on 50mm lens, macro details on the texture of the packaging. High-end department store holiday campaign vibe, 8k. ``` *** **Prompt 19 — The TikTok Shop Native** This prompt generates raw, vertical, "user-generated" style content perfect for TikTok, Reels, and Shorts. It intentionally degrades the "studio quality" to make the ad feel native to the feed, increasing hold rates. Use this for top-of-funnel awareness. **Input Variables** * \[Product]: The item being sold * \[Action]: How the user is interacting with it (holding it, applying it, wearing it) * \[Setting]: Where the video is being shot (messy bedroom, bright kitchen, gym car park) **The Prompt** ```text theme={null} vertical full screen smartphone photo posted to social media, POV shot, a hand holding [Product] in a [Setting], [Action], natural window lighting, raw unedited aesthetic, slight motion blur, shot on iPhone 14 Pro, user generated content style --ar 9:16 ``` *** **Prompt 20 — The Before/After Split** A classic direct response format reimagined with AI. This visualizes the transformation your product offers. Use this for problem-aware audiences who need visual proof of efficacy (skincare, cleaning, home organization). **Input Variables** * \[Problem State]: Visual description of the issue (ex: stained carpet, acne skin, cluttered desk) * \[Solved State]: Visual description of the result (ex: pristine carpet, glowing skin, organized desk) * \[Product]: The item that caused the change **The Prompt** ```text theme={null} split screen comparison image. Left side: [Problem State], dull lighting, desaturated tones. Right side: [Solved State], bright lighting, vibrant tones. In the center dividing line, a [Product] is superimposed. High definition advertising photography, photorealistic --ar 4:5 ``` *** **Prompt 21 — The How-It-Works Sequence** This prompt generates a triptych (three-panel) layout to explain a routine or assembly process. It reduces friction by showing the customer exactly what to do. Use for complex goods or multi-step routines. **Input Variables** * \[Step 1]: Visual of the first action * \[Step 2]: Visual of the second action * \[Step 3]: Visual of the final result * \[Style]: The aesthetic (minimalist vector, photorealistic studio, hand-drawn sketch) **The Prompt** ```text theme={null} three-panel horizontal instructional layout. Panel 1: [Step 1]. Panel 2: [Step 2]. Panel 3: [Step 3]. [Style] aesthetic, clean background, instructional design, sequential storytelling, high detail --ar 3:2 ``` *** **Prompt 22 — The Subscription Box Reveal** This simulates the dopamine hit of unboxing. It shows the value proposition of a bundle or subscription by displaying quantity and variety. Use for welcome offers or "build your own bundle" ads. **Input Variables** * \[Box Color/Branding]: Description of the packaging * \[Contents]: List of items inside * \[Surface]: Texture underneath the box (marble counter, wooden floor, bedspread) **The Prompt** ```text theme={null} top-down flat lay photography, a [Box Color/Branding] shipping box partially open on a [Surface], inside the box are [Contents] arranged neatly and overflowing slightly, soft studio lighting, 4k resolution, unboxing aesthetic --ar 4:5 ``` *** **Prompt 23 — The Detail Callout / Annotated Hero** This prompt creates a high-end "tech specs" visual. It isolates the product and uses graphical lines to point out specific value props. Essential for products where material quality or specific engineering features justify the price point. **Input Variables** * \[Product]: The item * \[Feature 1]: Visual detail to highlight * \[Feature 2]: Visual detail to highlight * \[Background]: Usually solid color or gradient to make text pop **The Prompt** ```text theme={null} macro product photography of [Product], extreme close-up on [Feature 1] and [Feature 2], thin white graphic lines pointing to specific product details, floating text placeholders, [Background] background, premium editorial tech style, sharp focus --ar 1:1 ``` *** **Prompt 24 — The Testimonial Card** This blends social proof directly into the creative. Instead of overlaying text in post-production that looks pasted on, this prompt attempts to integrate the "review" aesthetic into the scene (ex: a note, a screen, or a floating graphic). Use for retargeting. **Input Variables** * \[Product]: The item * \[Setting]: Lifestyle environment * \[Quote]: Short 3-5 word review snippet (ex: "Best Sleep Ever") **The Prompt** ```text theme={null} lifestyle photography of [Product] sitting on a table in a [Setting], a floating glass-morphism UI card appears next to the product with 5 gold stars and the text "[Quote]", depth of field, warm lighting, social proof ad creative --ar 4:5 ``` *** **Prompt 25 — The Retargeting / FOMO Creative** Designed to stop the scroll for cart abandoners. This prompt uses aggressive visual language—high contrast, bold colors, and "sale" aesthetics—to signal urgency. **Input Variables** * \[Product]: The item * \[Offer Text]: The deal (ex: 50% OFF, FLASH SALE) * \[Color Palette]: Usually red, yellow, or brand contrast colors **The Prompt** ```text theme={null} dynamic advertising shot of [Product], dramatic hard lighting, bold 3D typography integrated into the scene reading "[Offer Text]", [Color Palette] color scheme, motion lines, high energy, urgency aesthetic, commercial photography --ar 1:1 ``` *** # Part 6: Prompt Chaining & Batch Production You cannot scale DTC creative by treating every prompt as a one-off art project. To feed the Facebook and TikTok algorithms, you need volume. We achieve this through **Prompt Chaining** and the **30-Minute Batch Workflow**. **The Seed Strategy** AI generation is inherently random. If you find a generated image layout you love, you must lock the "Seed" (the random noise pattern used to generate the image). In NanoBanana 2, once you have a winning composition, keep the seed number constant and only change one variable at a time (ex: change the background color from "blue" to "red" or the model from "male" to "female"). This allows you to create consistent campaign assets rather than disjointed images. **The 30-Minute Batch Workflow** 1. **Define Core Angles (5 Minutes):** Don't just say "shoes." Define three distinct angles: *The Commuter* (shoes on subway), *The Runner* (shoes on track), and *The Studio* (floating product shot). 2. **Execution with Style Anchors (20 Minutes):** Use "Style Anchors" in your prompts—specific photographers, lighting setups (ex: "Rembrandt lighting"), or artistic styles—to ensure brand consistency. Run your prompts in batches of 10. 3. **The Cull (5 Minutes):** This is the most important step. You will generate garbage. That is part of the process. Apply the **Rule of 5**: For every 1 usable ad creative, you generally need to generate 5 options. Delete the failures ruthlessly. By chaining prompts (using the output of one prompt to inform the next) and batching production, you move from "prompt engineer" to "creative director," curating assets rather than struggling to build them pixel by pixel. *** # Part 7: The Quality Gate In the world of Generative AI, speed is easy; quality is the bottleneck. As we move from wellness-specific advice to broad DTC application, the "Quality Gate" becomes your primary defense against brand dilution. **The Allocation Matrix** Do not use AI for everything. * **Tier 1 (Brand Hero):** The main image on your homepage or packaging? **Hire a human photographer.** You need absolute control and 100% reality. * **Tier 2 (Performance Testing):** Facebook ad variants, email headers, blog thumbnails? **Use AI.** This is where volume wins. **The 3-Second Scroll Test** Before exporting any AI image, shrink it down to the size of a mobile phone screen and scroll past it quickly. Does the product read clearly? Is the focal point obvious? If the image looks cluttered or confusing at thumb-size, it fails the gate. **DTC Compliance Checklist** Unlike general art, commercial DTC creative has legal boundaries. 1. **Product Accuracy:** If your sneaker has 6 eyelets, the AI cannot render 8. If your serum is blue, the AI cannot make it purple. You are liable for "bait and switch" if the creative doesn't match the delivered goods. 2. **Text Accuracy:** NanoBanana 2 is better at text, but not perfect. Zoom in on every label. Ensure "Ingredients" doesn't spell "Ingrednet." 3. **Claim Visualization:** Be careful with "Before/After" generations. You cannot use AI to exaggerate results (ex: generating a person who lost 50lbs instantly). The visual claim must represent typical results. 4. **Platform Compliance:** Meta and TikTok have strict policies against "non-existent functionality" (ex: play buttons that don't play) or overly graphic zoomed-in body parts (common in skincare AI generations). **Closing the Loop** Scale what passes. Kill what doesn't. The goal was never to replace a photographer, it was to stop letting production costs decide which ideas get tested. Now they don't. # **If you want this built for your brand** This playbook is a starting point. A well-directed AI creative workflow, one that is mapped to your customer segments, integrated with your Klaviyo flows, and pressure-tested against your existing top performers, takes effort to build properly. If that is worth a conversation: [amlan@madebydas.com](mailto:amlan@madebydas.com) # The Operator Prompt: A System for Creating Content That Actually Compounds Source: https://intelligence.madebydas.com/playbooks/operator-prompt A six-part content creation system for DTC brand operators and CMOs — how to create content that builds institutional intelligence rather than disposable volume. Covers prompt architecture, context documents, review mining, and the creative brief workflow that separates content that compounds from content that resets. From DAS — The Audience Development Agency. # The Operator Prompt ## A System for Content That Compounds Most AI-generated content resets. It is produced for this week's campaign, consumed or ignored, and then next week's prompt starts from zero. The context disappears. The insights disappear. The institutional knowledge that should be compounding is instead evaporating at the end of every conversation. This guide is a system for the opposite. The Operator Prompt is a framework for DTC brand operators who want content creation to compound — where each piece builds on the intelligence from the last, where the AI has more context over time rather than less, and where the output of creative work feeds back into the customer intelligence that drives the next creative brief. *** ## Part 1: The Context Architecture The gap between generic AI output and genuinely useful output is context. The gap between useful output and compounding output is persistent context — context that grows over time rather than resetting with each conversation. Build five living documents. These are not documents you create once and file. They are documents you maintain and feed into every AI content session. ### Document 1: Brand DNA Everything a content creator needs to know about the brand in a single document that can be uploaded to any AI session. **Contents:** * Brand voice principles (3–5 rules, written as constraints — what the brand never says, not just what it says) * ICP definition — specific, behavioral, not demographic. Not "women 35–55" but "CMOs at \$25M brands who have been burned by agencies that prioritized vanity metrics" * Key messaging pillars — the 3–4 ideas the brand owns and returns to consistently * Vocabulary list — the specific words the brand uses and avoids * Tone examples — 2–3 samples of writing that represents the brand voice at its best **The test:** If someone unfamiliar with the brand reads this document and then writes a piece of content, would it sound like the brand? If not, add more specificity. ### Document 2: Customer Intelligence Brief The rolling weekly analysis of the customer file. This is the document built from the Claude agent workflow in the agents playbook — the Monday morning intelligence brief that accumulates over time. **Contents:** * Current RFM distribution (Champion segment size and revenue share) * Most recent cohort LTV findings * Current at-risk customer volume and revenue at stake * Key behavioral insights from the last 30 days * Any anomalies or trend shifts **Why it matters for content:** When the content team briefs creative, they should know which segment they are writing for. The Customer Intelligence Brief tells them who the Champion is, what the at-risk customer looks like, and where the file is growing or shrinking. Content that is not segment-aware is not intelligence-driven content. ### Document 3: Customer Voice Library The living repository of verbatim customer language. Built from review mining (see Agent 13 in the Claude Agents Playbook), support ticket analysis, post-purchase survey responses, and community engagement. **Contents:** * Top 15 phrases customers use to describe product transformation * Top 10 objections customers voice before purchasing * Top 5 pain points the product resolved (verbatim) * Language customers use when they are disappointed vs. delighted * Any phrases that appear 3+ times verbatim across reviews **The rule:** At least 3 verbatim customer phrases must appear in every creative brief. Not paraphrased. Verbatim. This is the single most reliable way to make creative resonate with the audience that matters. ### Document 4: Competitive Landscape Updated weekly (see Agent 17 in the Claude Agents Playbook). What competitors are saying, where they are weak, what customer needs they are not meeting. **The content application:** Every campaign brief should include one line: "What this says that competitors can't say." If it doesn't exist, the campaign is producing parity content. Parity content is expensive slop. ### Document 5: Performance Archive What has worked and what has not, written in enough detail to be useful. Not just metrics — context. **Contents:** * Best-performing campaigns with context (what was the segment, the offer, the format, the timing) * Worst-performing campaigns with diagnosis (why it did not work — wrong segment, wrong message, wrong timing, wrong platform) * Subject lines ranked by open rate and revenue per recipient * Ad creative ranked by ROAS and 90-day cohort LTV (not the same list) **The compounding effect:** When a content creator has access to a Performance Archive with 18 months of annotated history, they are not starting from zero. They are starting from the accumulated knowledge of everything the brand has tried and learned. *** ## Part 2: The Brief-First Workflow Most content creation starts with the output. Write the email. Write the ad. Write the post. Then figure out what it was supposed to accomplish. The Operator Prompt workflow starts with the brief. The brief is the intelligence layer. The output is downstream. ### The Brief Template Every content piece — email, ad, social post, landing page — should be preceded by a brief that answers these questions: **Who is this for?** (RFM segment, not demographic) **What do they already believe?** (What is their current understanding of the product or offer — informed by the Customer Voice Library) **What do we want them to do?** (One action. Not "engage with the content" — a specific action with a specific outcome.) **What is the one insight this piece of content will deliver?** (The thing they did not know before that makes the action logical. If there is no insight, there is no reason to produce the content.) **What customer language must appear verbatim?** (3+ phrases from the Customer Voice Library) **What is this content's relationship to the customer file?** (Is this a Champion retention piece? A second-purchase driver for Recent customers? A winback for At Risk? If the answer is "general audience," reconsider whether to produce it.) *** ## Part 3: The Prompting Framework The difference between prompts that produce usable content and prompts that produce generic output is specificity of context. **The template:** ``` You are writing [content type] for [RFM segment]. Context documents I'm providing: - Brand DNA (uploaded) - Customer Voice Library (uploaded) - Campaign brief (below) Campaign brief: - Objective: [specific action] - Segment: [Champion / Loyal / Recent / At Risk] - Insight to deliver: [one sentence] - Customer phrases to include verbatim: [3 phrases from Customer Voice Library] - Offer or CTA: [specific] - Platform/format: [email, Meta feed, SMS, etc.] - Tone constraint: [reference Brand DNA voice rules] Do not use: - Generic DTC opening lines ("We noticed you haven't purchased recently") - Discount-leading copy for Champion segments - Industry jargon not found in the Customer Voice Library - Any phrase that could appear in a competitor's content Output: [specific format — email with subject line and preview text, or 3 ad headlines with primary text] ``` The specificity of the brief determines the quality of the output. Garbage in, garbage out is true for AI as it is for every other creative process. The Operator Prompt is a system for ensuring the context going into the AI session is never garbage. *** ## Part 4: The Feedback Loop Content that compounds has a feedback loop built in. After every campaign: **What went into the Performance Archive:** * What the brief was (not just the creative) * Which segment received it * The specific customer phrases that were used * What the open rate, click rate, and revenue per recipient were * What the 30-day repeat purchase rate was for customers who received it * The diagnosis: why did it perform or underperform? **What gets updated in the Customer Voice Library:** * Any new phrases from replies, replies, or support conversations that the campaign triggered * Any objections that surfaced in customer service after the campaign * Any language that generated outsized engagement Over 12 months, this feedback loop produces an institutional knowledge base that no competitor without the same system can replicate. That is the compounding. Not more content — smarter context. *** ## Part 5: The Anti-Patterns Five content behaviors that signal a team is producing volume rather than intelligence: **The generic win-back.** "We miss you!" sent to every lapsed customer regardless of value, purchase history, or behavioral segment. This is not a retention strategy. It is a list management activity. **The discount-first Champion email.** Sending a 20% off code to a Champion-segment customer who would have purchased without it. This is not generosity. It is margin destruction. **The campaign that forgets the brief.** Content that is well-crafted but has no clear connection to a specific behavioral segment or business outcome. Beautiful slop. **The A/B test without a hypothesis.** "Let's test two subject lines and see which performs better." Without a specific hypothesis about why one would outperform the other — and what that result would tell us — the test produces data without intelligence. **The reset.** Starting each campaign from zero without referencing the Performance Archive, the Customer Voice Library, or the Customer Intelligence Brief. Each piece of content should be informed by everything that came before it. *** ## Part 6: The Minimum Viable System For teams that do not have the capacity to implement the full framework immediately, the minimum viable version: 1. **One document:** Build the Customer Voice Library first. Extract verbatim phrases from reviews. Require 3 per brief. This change alone will produce measurably different creative output within 30 days. 2. **One rule:** Every email goes to a segment, not a list. Define the segment before you write the email. 3. **One metric:** Track revenue per recipient by segment, not open rate. The segment that receives the right message will show it in revenue per recipient first. Build from there. Click **Open in Claude** above to start building your Brand DNA document and Customer Voice Library with Claude's help. Upload your last 100 customer reviews and Claude will extract the verbatim phrase library that drives the rest of the system. # Playbooks Lander Source: https://intelligence.madebydas.com/playbooks/playbooks-lander # The Voice-to-Agent Playbook Source: https://intelligence.madebydas.com/playbooks/voice-to-agent The exact Wispr Flow, Claude Cowork, and Granola setup that replaced 60 hours of sales, marketing, and ops workflows. A three-tool voice-to-agent system. ## The Exact Wispr Flow + Claude Cowork + Granola Setup That Replaced 60 Hours of Sales, Marketing, and Ops Workflows A three-tool voice-to-agent playbook to deploy this week. *** | # | Section | Description | | - | ---------------------------- | -------------------------------------------------- | | 1 | **Part 1: The Problem** | You're Still Typing Into Chatbots | | 2 | **Part 2: The Stack** | Wispr Flow + Claude Cowork + Granola | | 3 | **Part 3: The Setup** | 10-Minute Configuration + CLAUDE.md Trigger System | | 4 | **Part 4: The 8 Workflows** | Voice-to-Agent Delegation | | 5 | **Part 5: Mobile Workflow** | Delegate From Anywhere | | 6 | **Part 6: Pro Tips + Recap** | Common Mistakes and Final Words | *** *** > **Note: Before You Dive In** > > This playbook is dense by design. It is meant to be returned to, not skimmed once. Bookmark it, build with it, come back when you are stuck. > > It was built specifically for DTC, retail, and ecommerce operators, but the stack works for any brand competing for attention in a content-saturated and AI-volume market. If you run campaigns, manage creative, or make decisions about what gets produced and when, this is for you. > > If you want this built for your brand or want a second set of eyes on your current content system, reach out to Amlan directly -- [amlan@madebydas.com](mailto:amlan@madebydas.com) *** ## Part 1: The Problem -- You're Still Typing Into Chatbots You pay for AI. You type into a chat window. You wait for a response. Then you copy that text and paste it into a Doc or an email. Congratulations, you just invented manual labor with a subscription fee. Most people use Claude like a search engine. They treat it like a slot machine: type a prompt, pull the lever, hope for a good result. That is the "Chatbot" era, and it is dead. The average person types 40 WPM. Your brain moves at 800+ WPM. That means 95% of your output dies in the gap between your brain and your keyboard. Your keyboard is a bottleneck, it was designed in 1873 to slow typists down, and we are still using it to talk to the most advanced technology in history. I hit a wall last Tuesday at 4:00 PM. I had to write a project brief that required context from three different files. I started typing the prompt: "Claude, look at the project notes. Then look at the budget. Then draft a..." My fingers felt FAT. I realized I was spending more time explaining the task than just doing it. By the time I finished typing, I forgot the third point I wanted to make. The "Prompt Fatigue" is real, and it stops you from doing high-level work. I identified three traps keeping you stuck: ### 1. The Copy-Paste Tax You prompt. You copy. You paste. No file access. No memory. No autonomy. You are the "middleware", you sit between the AI and your work. If you are the one moving the data, the AI isn't working for you. You are working for the AI. ### 2. The Keyboard Bottleneck Complex prompts take 5 minutes to type. In that time, you lose the spark. You get tired of typing, so you "short-hand" your instructions. The result? Lazy prompts lead to bad output. You get "average" because you were too tired to be specific. ### 3. The Serial Processing Trap One prompt. One response. One task. It's linear. Meanwhile, your to-do list has 15 items that could all run at the same time. You are babysitting a chatbot when you should be managing a department. ### 4. The Meeting Memory Gap Your best strategic context doesn't live in files, and the best context from calls, syncs, and strategy sessions never makes it to the brief. Nuance gets lost, and you end up working from a diluted version of what was actually said. It's not that you're not paying attention, it's that there's no system connecting what happens in your meetings to what your AI executes. Your agent is only as smart as the context it receives -- and if that context stops at your file folder, you're leaving the best inputs out of the loop entirely. ### The Better Way Imagine speaking a task in plain English while walking to get coffee. An agent reads your local files, spins up sub-agents to do research, writes the final work, and saves it to your hard drive. No typing. No copy-pasting. No babysitting. This is not a "future" vision. This is the stack I use today. It's called **Voice-to-Agent Delegation**, and it turns "typing prompts" into "managing a workforce." Here is how the stack works. *** ## Part 2: The Stack -- Wispr Flow + Claude Cowork + Granola To kill the keyboard, you need two tools that work in sync. One handles the input (your voice), and the other handles the execution (autonomous work). Together, they create a new category of productivity that makes typing feel obsolete. ### Tool 1: Wispr Flow (The Input Engine) [Wispr Flow](http://wisprflow.ai/) is not "dictation." It is not Siri. It is a contextual voice AI that lives on your Mac, Windows, and iPhone. It solves the "Input Problem" with raw power: * **500ms Latency:** It types as fast as you speak. There is no lag between your thought and the screen. * **85% Zero-Edit Rate:** You don't have to fix typos. What it writes is clean enough to send. * **3x More Accurate:** It understands jargon, messy grammar, and industry-specific vocabulary. * **4x Faster Than Typing:** You speak at 180 WPM. Your fingers top out at 40. It doesn't just transcribe, it formats. It reads your screen. It knows the context of your work. It learns your vocabulary over time. If I say "Claude Cowork," it spells it right. It doesn't type "Cloud Co-work." This is why teams at Meta and Klarna use it. It's the first voice tool that is genuinely faster than a keyboard. I used to use the built-in Mac dictation. It was a nightmare, it would miss every third word, and I spent more time fixing the text than typing it. Wispr Flow feels like a direct link from your brain to the screen. ### Tool 2: Claude Opus 4.6 Cowork (The Execution Engine) Claude Cowork is not a chatbot. It is an agent with **Direct File Access.** It can read and write files on your laptop, spin up sub-agents for research, use a browser for live web data, do multi-step tasks without you watching, and use `CLAUDE.md` to remember your style across sessions. Think of it as a junior employee who has access to your folders. You don't "chat" with it, you delegate to it. ### Tool 3: Granola (The Meeting Intelligence Layer) [Granola](http://granola.ai) runs in the background of every meeting, on your Mac or iPhone, and produces a structured, searchable note by the time the call ends. No manual note-taking or transcription cleanup or trying to remember what was said while you're already in the next meeting. Wispr Flow handles input. Claude handles execution. But neither of them knows what was said in your last client call, your weekly team sync, or the strategy session where your best insight of the quarter surfaced. *That's where Granola comes in.* Granola connects natively to Claude via the Claude connector to now have access to your meeting history. When you delegate a task, Claude can pull context from your actual calls without you re-explaining anything. This closes the Meeting Memory Gap entirely. The insight from the call becomes an input to the agent. ### The Combination: A New Category of Work Granola handles the **MEMORY** (meeting intelligence, stored context Claude can access natively). Wispr Flow handles the **INPUT** (your voice, clean instructions). Claude Cowork handles the **EXECUTION** (instructions + context, autonomous work, files). Together: your meetings feed the agent, you speak the instruction, the work lands in your folder. You are no longer a "Prompter." You are no longer the middleware between your calls and your deliverables. You are an ROI Operator. Here is a micro-workflow to illustrate the power: ```text theme={null} Voice Command: "Pull my Granola notes from this week's account review. Cross-reference with the CRM export in the inbox folder. Find every account with a deliverable flagged but no activity in the last 7 days. Draft a one-paragraph status update for each. Save to the output folder." ``` Wispr Flow types that into Claude in under 1 second. Claude pulls your meeting notes via the Granola connector, reads the CRM export, matches the two, writes the status updates, and saves the file. Total keyboard input: zero. Total context reconstructed from memory: **zero**. Total time saved: **3 hours.** This works for every department: Sales, Marketing, Ops, Product, Finance, Strategy, and HR. Wispr Flow works everywhere: Slack, Email, WhatsApp, Notion. It has a **Whisper Mode** so you can use it in a coffee shop without looking crazy. Granola runs silently in the background of every meeting on Mac and iPhone. The tech has finally caught up to your brain. *** ## Part 3: The Setup -- 10-Minute Configuration You don't need a degree in AI to set this up. You need 10 minutes and coffee. Here is how you wire your brain to your hard drive. ### Step 1: Wispr Flow (3 Minutes) Go to [wisprflow.ai](http://wisprflow.ai) and download the app. Once installed, set your activation hotkey, I use **Option + Space.** It becomes your "Listen to Me" button. Open the **Personal Dictionary** and add 15 terms you use every day: company names, client names, jargon, acronyms. Wispr Flow learns these immediately and will never misspell them. Turn on **Whisper Mode**, this lets you dictate in a library, on a flight, or in an open office. The AI picks up quiet speech. Nobody hears you, and the work gets done. ### Step 2: Granola (2 Minutes) Download [Granola](http://granola.ai) on your Mac or iPhone. It runs silently in the background of every meeting, no setup required per call, no manual triggers. By the time a meeting ends, the note is already written. The critical setup step: connect Granola to Claude via the Claude connector. This gives Claude native access to your full meeting history. You don't copy and paste. You don't re-explain what was said on a call. Claude already has it. To connect: open Claude settings, navigate to Connectors, and enable Granola. Once connected, any voice command you give Claude can reference your meeting history directly. ### Step 3: Claude Cowork (5 Minutes) Open the Claude Desktop App and switch to Cowork mode. Give it a home base on your machine -- I use a folder called "Agent Workspace." Inside that folder, build this structure: ```text theme={null} ~/cowork/ inbox/ <- Drop raw files here for processing output/ <- Where Claude saves finished deliverables templates/ <- Your style guides and reusable frameworks archive/ <- Where old files go after processing ``` ### Step 4: Building Your CLAUDE.md: The Trigger System Create a file named `CLAUDE.md` in the root folder. Most guides will tell you to use this for role definition and preferences. That's a start, but the real unlock is using it to define trigger phrases that execute full workflows from a single short voice command. **Here's why this matters:** if you have to speak a 60-word research prompt every time you want a signal sweep, you've recreated the keyboard bottleneck with your voice. The goal is to speak six words and have Claude execute the full defined workflow automatically. **Here's how this works:** Claude reads `CLAUDE.md` at the start of every session. When you speak a trigger phrase via Wispr Flow, Claude matches it to the defined workflow and executes the full instruction set, file access, research parameters, output format, save location, without you specifying any of it. Here is the full `CLAUDE.md` template to get you started: ```markdown theme={null} # ROLE AND CONTEXT I am [Your Name/Role] at [Company]. We work with [brief description of clients/industry]. My team is [X] people. We run [X] client accounts simultaneously. # OUTPUT PREFERENCES - Format: Markdown unless specified otherwise - Length: Concise. No padding. Lead with the output. - Tone: Direct, professional, no filler phrases - File naming: [ClientName]-[TaskType]-[YYYY-MM-DD] # FILE RULES - Never delete from inbox/, always copy to archive/ - Save all deliverables to output/ unless instructed otherwise - Confirm file save location before closing task # TRIGGER PHRASES Define your workflow triggers below. Each trigger phrase you speak via Wispr Flow maps to the full instruction set underneath it. Claude reads this file at the start of every session. You speak six words. Claude executes sixty. Use this format for each trigger: ## [TRIGGER NAME] When I say "[short trigger phrase]", execute the following: [Full workflow instruction] [Output format] [Save location] ``` Each workflow in Part 4 includes the exact trigger phrase block to copy into this section. Build your CLAUDE.md as you work through the workflows -- add each trigger once, and it's available every session from that point forward. **One critical setting:** Web search must be enabled in Claude Cowork for any trigger that requires live data -- signal sweeps, competitor research, category intelligence. This is a toggle in settings and is not on by default. Without it, Claude pulls from training data, not current content. Turn it on before you run your first sweep. ### Step 5: The Connection (2 Minutes) There is no complex API integration. Open Claude Cowork, hit your Wispr Flow hotkey, and start talking. Wispr Flow streams your voice into Claude's input field. Claude reads your `CLAUDE.md`, accesses your Granola meeting history via the connector, reads your local files, and executes. The full loop: ```text theme={null} Granola captures meeting context | Wispr Flow carries your voice instruction | Claude executes with access to both ``` Total keyboard input: zero. Total context reconstructed from memory: zero. You are no longer a Prompter. You are no longer the middleware between your meetings and your work. You are an ROI Operator. *** ## Part 4: The 7 Voice-to-Agent Workflows This is where the ROI happens. Each workflow below includes two things: the trigger phrase definition to copy into your CLAUDE.md once, and the short voice command you speak via Wispr Flow to fire it. Set up the trigger once. Speak two to eight words every time after that. ### 1. Sales -- The "Pipeline Briefer" Add this to your CLAUDE.md once: ```text theme={null} ## PIPELINE BRIEFER When I say "run the pipeline briefer", execute the following: Read the CRM export in inbox/. Find every deal with no activity in the last 14 days. For each stalled deal, pull any Granola meeting notes from the last 30 days that reference that account or contact name. Draft a re-engagement email for each deal that: - References the last recorded pain point from the CRM - Incorporates any specific context or language from Granola call notes if available - Sounds human, not templated Save all drafts to output/ as pipeline-outreach-[date].md ``` **Voice command:** `Run the pipeline briefer` Claude checks the CRM export for stalled deals, then cross-references your Granola meeting history for any call context on those accounts. The re-engagement emails reference actual conversations, not just CRM fields. The difference between an email that gets opened and one that gets deleted is usually one specific detail that proves you were paying attention. You skip the manual CRM filter entirely and jump straight to outreach. **Deliverable**: `pipeline-outreach-[date].md` **Time Saved:** 2.5 hours to 5 minutes. ### 2. Marketing -- The "Content Repurposer" Add this to your CLAUDE.md once: ```text theme={null} ## CONTENT REPURPOSE When I say "repurpose [filename] for LinkedIn", execute the following: Read [filename] from inbox/. Pull my style guide from templates/. Create 5 LinkedIn posts with these angles: contrarian take, data post, how-to guide, story, and list. Formatting rules: short paragraphs, punchy hooks, arrows for bullets, no hashtags. Save to output/ as linkedin-bundle-[filename]-[date].md ``` **Voice command:** `Repurpose "Future of Voice AI" for LinkedIn` Repurposing is the most time-consuming task in marketing, it kills creative energy without producing anything new. Claude pulls your style guide from the templates folder, applies your formatting rules, and splits one long idea into five sharp hooks with different angles. You get a week of LinkedIn content from a single asset. **Deliverable:** `linkedin-bundle-[filename]-[date].md` **Time Saved:** 3 hours to 6 minutes. ### 3. Ops -- The "SOP Auditor" Add this to your CLAUDE.md once: ```text theme={null} ## SOP AUDIT When I say "run the SOP audit", execute the following: Scan every document in the SOPs folder in inbox/. Look for: overlapping processes, outdated steps, and manual steps that could be automated with Zapier or Make.com. Produce a prioritized improvement table with columns: Process Name | Issue Type | Recommended Fix | Automation Potential (High/Med/Low) Save to output/ as sop-audit-[date].md ## SOP PULL When I say "run a Granola SOP pull for [process]", execute the following: Search Granola meeting notes for every instance where [process] was discussed — how it works, who owns it, any issues flagged. Synthesize into a first draft SOP. Flag any steps where descriptions conflicted across different meetings. Save to output/ as sop-draft-[process]-[date].md ``` **Voice commands:** `Run the SOP audit` `Run a Granola SOP pull for client onboarding` Two triggers, two different use cases. The first audits your existing documented SOPs for gaps and automation opportunities. The second pulls every time a process was discussed across your entire Granola meeting history and builds a first draft from it. Most SOPs fail because the person writing them wasn't in all the relevant conversations. The Granola pull fixes that, the draft arrives pre-loaded with the institutional knowledge that usually gets lost between the idea and the document. **Deliverable**: `sop-audit-[date].md` or `sop-draft-[process]-[date].md` **Time Saved:** A full day of auditing to 15 minutes. ### 4. Product -- The "Feedback Synthesizer" Add this to your CLAUDE.md once: ```text theme={null} ## FEEDBACK SYNTHESIZER When I say "run the feedback synthesizer", execute the following: Read the customer feedback CSV in inbox/. Cluster into the top 5 complaint types. For each cluster provide: - Count of mentions - 3 representative quotes - One-paragraph recommended fix Pull any Granola meeting notes where these complaint types were discussed by the team and append relevant context to each cluster. Format the full output as a product brief. Save to output/ as feedback-brief-[date].md ``` **Voice command:** `Run the feedback synthesizer` Reading 1,000 rows of customer feedback manually is the kind of work that kills momentum. Claude does the thematic clustering, surfaces the voice-of-customer patterns, and cross-references your team's own meeting discussions about those issues. The brief arrives ready to drop into your next product meeting, with both the raw data and the internal context already synthesized together. **Deliverable**: `feedback-brief-[date].md` **Time Saved:** 4 hours of data crunching to 8 minutes. ### 5. Finance -- The "Expense Cruncher" Add this to your CLAUDE.md once: ```text theme={null} ## EXPENSE CRUNCHER When I say "run the expense cruncher", execute the following: Read all receipt images and the bank export in inbox/. Use Vision to match each receipt image to its corresponding bank row. Categorize every expense by: software, travel, meals, contractors, and other. Calculate totals for each category. Identify which category grew most since last month. Flag any single expense over $500 for review. Save to output/ as expense-summary-[month]-[date].md ``` **Voice command:** `Run the expense cruncher` Claude uses Vision to read the receipt images, matches each one to the corresponding bank row, does the math, and surfaces the budget leak. You get a formatted expense report with category totals, month-over-month trend analysis, and a flagged review list, without opening a single spreadsheet. **Deliverable**: `expense-summary-[month]-[date].md` **Time Saved:** 90 minutes of spreadsheet work to 5 minutes. ### 6. Strategy -- The "Competitor Scout" Add this to your CLAUDE.md once: ```text theme={null} ## COMPETITOR SCOUT When I say "run a competitor scout for [Company A], [Company B], [Company C]", execute the following: Use the browser to find current pricing pages for each company. Find any new features, product updates, or positioning changes from the last 90 days. Pull any Granola meeting notes where these competitors were mentioned and include our team's existing perspective. Build a battlecard with: - Feature comparison table - Pricing comparison - Recent moves by each competitor - A "Where We Win" section with specific talking points Save to output/ as battlecard-[date].md ``` **Voice command:** `Run a competitor scout for [Company A], [Company B], [Company C].` Competitive research done manually takes half a day and still misses things. Claude scrapes three sites simultaneously, surfaces the hidden pricing pages, and incorporates your team's existing perspective from past meeting notes. The battlecard arrives sales-ready. The "Where We Win" section gives your team ready-made talking points before the next pitch. **Deliverable**: `battlecard-[date].md` **Time Saved:** 3 hours of research to 12 minutes. ### 7. HR -- The "Resume Radar" Add this to your CLAUDE.md once: ```text theme={null} ## RESUME RADAR When I say "run the resume radar for [role]", execute the following: Read all PDF resumes in inbox/. Pull scoring criteria from the job description in templates/ if available. If not, apply general criteria for the role type. For each candidate: - Score 1-10 against criteria - Write a two-sentence summary of fit - Flag any standout skills or experience Produce a ranked shortlist of the top 5. Save to output/ as candidate-shortlist-[role]-[date].md ``` **Voice command:** `Run the resume radar for Senior Developer` Skimming resumes is the highest-cost, lowest-leverage use of any hiring manager's time. Claude opens all the PDFs, extracts the relevant experience from each one, applies your scoring criteria, and surfaces the top five with a summary of why each made the cut. You skip straight to the conversations that matter. Deliverable: `candidate-shortlist-[role]-[date].md` **Time Saved:** 2 hours of skimming to 10 minutes. ### 8. Operations -- The "Weekly Sync-to-Action" Add this to your CLAUDE.md once: ```text theme={null} ## WEEKLY DEBRIEF When I say "run the weekly debrief", execute the following: Pull this week's Granola meeting notes via the connector. Identify and organize by: 1. Account priorities ranked by urgency 2. Deliverables flagged in meetings with no update 3. Open decisions that need a response this week 4. Team assignments with no activity since assigned Format as a prioritized action list organized by team member. Save to output/ as weekly-debrief-[date].md ``` **Voice command:** `Run the weekly debrief` That's the entire command. Claude pulls your Granola meeting notes from the week via the connector, maps every account by urgency, surfaces deliverables at risk, flags open decisions, and builds a prioritized action list sorted by team member. Your team starts Monday already aligned, no recap email written, no follow-up Slack thread, no "can someone send the notes from Thursday?" The 45 minutes of manual note consolidation that used to happen after your planning call now happens while you're still saying goodbye on the Zoom. **Deliverable**: `weekly-debrief-[date].md` **Time Saved:** 45 minutes of manual synthesis to under 2 minutes. *** ## Part 5: The Mobile Workflow -- Delegate From Anywhere The best ideas don't hit at your desk. They hit when you walk, when you commute, and when you wait for coffee. Up until now, mobile was a bottleneck -- Siri is a toy, and built-in dictation is a mess of typos. Wispr Flow on mobile changes this. It brings the same 500ms speed to your pocket. It isn't an app you open, it's a bubble on your screen. Tap, talk, done. ### The "Walking Delegation" Walking my dog in the morning and an idea hits: "I need to know how our pricing compares to those 3 new challenger brands for my client." I tap the Wispr Flow bubble, dictate the full task, and send it to my Claude session. By the time I get home, the competitive breakdown is sitting in my output folder. My machine worked while I walked. ### The "Gym Sprint" Between sets at the gym, I remember a client needs an update. I tap the bubble and say: "Draft a status update for the \[Client Name] project. Mention we hit the Phase 2 milestone. Ask for the feedback on the audience analysis. Save it as an email draft." No typing with sweaty thumbs. The task is queued and waiting for me. ### The "Coffee Shop Sprint" 8:00 AM -- Walk into the shop. 8:05 AM -- Dictate 3 tasks via Wispr Flow on my phone: a content repurpose job, an SOP audit, and a resume screen. 8:20 AM -- Check the output folder from my phone. Three deliverables are done. Total typing: zero. **One Setup Requirement for Mobile to Work This Way** The reason a short voice command can trigger a full research sprint from your phone is because the work of defining that workflow already happened, in your `CLAUDE.md` file. The six words you speak are a trigger phrase that maps to a complete instruction set Claude reads before every session. If you haven't built your `CLAUDE.md` triggers yet, the mobile workflow still works, you just have to speak the full prompt rather than the shorthand. Start there. Once you've run a few workflows manually via voice, you'll know exactly which ones are worth turning into triggers. Then go back to Part 3 and define them. If you want to move fast, try the **"10,000 Steps" Challenge.** Dictate every task you'd normally do in your first hour at the desk while you walk. By the time you hit 10,000 steps, your AI workforce has cleared your morning backlog before you even sit down. *** ## Part 6: Pro Tips + Common Mistakes I spent 40 hours breaking this system so you don't have to. If you want great output, follow these five rules. ### The 5 Rules for Voice Delegation **1. The "Definition of Done" Rule** Never say "help me organize." Instead, say: "Create a sheet with columns X, Y, and Z. Sort by revenue. Save it as Q1-Report.xlsx." Specificity kills errors. The more precise your instruction, the better Claude's output. **2. The "Context Dump" Technique** Spend 10 seconds on context before giving the task. If you're working from scratch, state it directly: "I am prepping for a board meeting. The audience is non-tech. Keep the tone executive." This single sentence changes the output dramatically, it sets the brain the AI uses for the entire session. If the context already exists in a meeting you've had, skip the reconstruction entirely. Say: "Pull my Granola notes from the board prep session last Tuesday and use that as your context for this task." Claude accesses your meeting history natively via the connector -- the specific language your client used, the direction your team agreed on, the nuance that made the insight sharp -- and uses it as the brief. Either way, the rule is the same: always set the context before the task. The output is only as specific as the input you give it. **3. Voice-Correct in Real Time** If Claude starts drifting, don't type a correction. Just speak: "Stop. That's not it. I need X, not Y." Wispr Flow types your correction directly into the chat field. You stay in voice mode the entire time. **4. The "Rule of 5"** Run 5 tasks per session, then start a new one. After 5 complex delegations, Claude's context window gets crowded. A fresh session keeps the agent sharp and prevents drift. **5. Use Whisper Mode for Sensitive Tasks** In open offices, coffee shops, or shared spaces, Wispr Flow's Whisper Mode picks up barely audible speech with full accuracy. Nobody around you hears what you're delegating. ### When NOT to Use Voice * **Typing passwords** or sensitive credentials (always use the keyboard). * **Environments too loud** even for noise cancellation (construction sites, concerts). * **Precise code syntax** (use voice for the logic and description, use the keyboard for the brackets and semicolons). *** ## Recap: What You Just Got You now have the complete blueprint to close the gap between your meetings, your voice, and your work. You have the **Memory Layer** (Granola) to capture meeting intelligence and feed it directly to your agent via the connector. You have the **Input Engine** (Wispr Flow) to turn your voice into clean instructions in under a second. You have the **Execution Engine** (Claude Cowork) to turn those instructions into autonomous work with full file access. You have the **CLAUDE.md Trigger System** to turn any complex workflow into a six-word voice command. And you have 8 copy-paste trigger phrases ready to deploy for Sales, Marketing, Ops, Product, Finance, Strategy, HR, and weekly operations. The result: you stop being the middleware between your meetings, your ideas, and your work. The context from every call feeds the agent. The instruction takes three seconds to speak. The deliverable lands in your folder. The gap between your brain and your work is gone. **Now go set it up and start delegating.** > **Note:** If you want assistance with this, or custom signal workflows, dashboard setup, or a full production sprint -- reach out. > > [**amlan@madebydas.com**](mailto:amlan@madebydas.com) / [newbiz@madebydas.com](mailto:newbiz@madebydas.com) # Buck Mason Brand Teardown: Fabric-First UX and Anti-Discount Positioning Source: https://intelligence.madebydas.com/teardowns/buck-mason A DTC brand teardown of Buck Mason's digital strategy — fabric-first product organization, anti-discount pricing philosophy, proprietary naming conventions, and editorial-first collaboration pages. Five specific tactical takeaways for DTC brand leaders. Part of the DAS brand teardown series analyzing what high-performing DTC brands do differently. # Buck Mason Brand Teardown ## Five Frameworks Every DTC Brand Can Steal Buck Mason breaks every conventional rule of DTC brand building. They do not run seasonal sales. They do not organize their product catalog the way every other apparel brand does. They name their fabrics the way luxury watchmakers name movements. They have built what looks like a traditional men's apparel brand but operates more like a materials company with a retail interface. In a market where DTC apparel brands compete on price and trend, Buck Mason competes on substance. Understanding how they built that position reveals transferable frameworks for any brand trying to stop competing on price. *** ## Framework 1: Fabric-First Information Architecture Most apparel brands organize their product catalog by category: shirts, pants, outerwear, accessories. This is logical from a merchandising standpoint and invisible as a brand differentiator. Every brand does it. The navigation tells the customer nothing about what makes the brand different. Buck Mason organizes primarily by fabric. Before you get to category, you encounter materials: Supima cotton, Pima jersey, ventile, slub linen. The product architecture makes the material the hero before the silhouette. **Why it works:** It creates a discovery experience that is inherently educational. A customer who did not know what Supima cotton was when they arrived knows what it is by the time they have navigated two pages. That education creates preference. When they understand why Supima cotton is better, they are not comparing a \$95 t-shirt to a \$35 t-shirt — they are comparing materials. The price difference becomes legible. **The underlying mechanism:** Information architecture is brand positioning. Where you start the customer's journey is a statement about what the brand believes matters. Buck Mason's navigation says: the fabric is the point. Everything else is downstream from that. **How to apply this:** Map your product catalog's current information hierarchy. What does the navigation prioritize? Is that the same thing your brand says it prioritizes in positioning copy? If the navigation prioritizes categories and the marketing prioritizes materials, ingredients, or process — that is a positioning gap. Solve it in the navigation, not in the ad copy. *** ## Framework 2: Anti-Discount Pricing Philosophy Buck Mason does not run sales. Not at BFCM. Not at end-of-season. Not as a customer acquisition tactic. The brand's discount philosophy is consistent and explicit: the price is the price. This is not a minor operational decision. It is an identity claim. **The margin economics:** A brand that runs 20% off sales twice a year trains its customer file to wait for those windows. The Champion-segment customers — the ones who would pay full price because they understand the product — learn to hold off. The discount attracts price-sensitive customers who churn at a higher rate. Over time, the promotional calendar becomes the business model. Revenue concentrates around sale periods. Full-price sell-through declines. The brand becomes dependent on discounts to hit revenue targets that require discounts to hit. Buck Mason exits this loop entirely. By never running sales, they never train customers to wait. Full-price sell-through is the only sell-through. The economics of the business do not have a promotional dependency. **The customer quality implication:** A customer base trained to pay full price has different behavioral characteristics than a customer base trained to buy on discount. Full-price customers have higher LTV, lower churn rate, and lower acquisition cost through word-of-mouth. They buy because they value the product — which means they keep buying when the product continues to deliver value. **How to apply this:** You do not have to eliminate discounts entirely. But you should run the analysis: what percentage of your Champion segment's purchases were made with a discount code? If the answer is more than 20%, you are discounting your best customers — the ones who would have bought at full price. The discount is margin destruction dressed as retention. *** ## Framework 3: Proprietary Naming as Differentiation Architecture Buck Mason names its fabrics as if they were proprietary ingredients. "Supima" is a real cotton designation, but Buck Mason treats it with the gravity of a pharmaceutical compound. The "Vintage Slub" is their name for a specific fabric texture. "Ventile" is a technical specification. The naming system makes commodity materials feel proprietary. **Why naming matters:** In a market where every brand can source similar materials, proprietary naming is one of the few forms of product differentiation that cannot be directly copied. If a competitor sources Supima cotton and calls it "premium cotton," the customer has no way to compare. If Buck Mason calls it Supima and builds educational content around what makes Supima different, they own the customer's understanding of that category. This is the same mechanism luxury watchmakers use to name their in-house movements, or the same way skincare brands name their hero ingredients. The name does not change the material — it creates ownership of the customer's understanding of the material. **The SEO and LLM implication:** A proprietary naming system creates search terms that only you own. No competitor can rank for "Buck Mason Vintage Slub" because that term is theirs. In an AI commerce world where product descriptions are consumed by AI agents, proprietary naming creates a differentiation layer that survives commoditized product data. **How to apply this:** Audit your product naming. Are you using category-generic descriptors ("premium cotton," "luxe fabric," "soft blend") or names that only your brand owns? The former is invisible in a competitive landscape. The latter is a moat. Invest in developing proprietary names for your hero materials, ingredients, or processes — then build content architecture around them. *** ## Framework 4: Editorial Collaboration Pages Buck Mason's collaboration pages read less like promotional content and less like standard partnership marketing. They are written the way a long-form editorial would be written — the story of the collaboration, the craft decisions, the specific details that make this object different. A typical DTC brand collaboration page: a photo of two logos, a product grid, some promotional copy. A Buck Mason collaboration page: the origin story of the relationship, the specific material or construction decisions that resulted from the collaboration, the point of view both parties brought to the project, and product copy that would hold up in a magazine feature. **The content signal this sends:** When the collaboration page is written with editorial care, it signals that the brand takes the collaboration seriously enough to invest in explaining it. That signal compounds. Customers who read the editorial context understand the collaboration differently than customers who see a co-branded product drop. They understand why the collaboration exists, not just that it does. **The SEO and GEO benefit:** Long-form editorial content on collaboration pages creates indexed content that AI agents can cite and search can surface. A brand teardown that mentions "Buck Mason collaboration with \[partner]" and links to the editorial page drives qualified traffic that generic collaboration marketing cannot. **How to apply this:** For your next collaboration, write the story before you design the product page. What is the specific reason these two brands or people are working together? What decision did that collaboration enable that neither party could have made alone? Write that story in 400–600 words and build the product page around it. *** ## Framework 5: The Retail Integration as Brand Expression Buck Mason's physical retail stores are not acquisition channels with different unit economics. They are three-dimensional expressions of the same brand system that drives the digital experience. The material the stores are built from — concrete, raw wood, industrial fixtures — communicates the same thing the fabric-first navigation communicates: substance over decoration. The in-store experience is coherent with the product philosophy. A customer who encounters Buck Mason in store and online encounters the same brand. This coherence is rarer than it sounds. Most brands treat physical retail as a separate channel with a separate customer experience. The result is a brand that feels different in different contexts — not because the product changes but because the environment around the product is inconsistent. **The retention implication:** Brand coherence across touchpoints is a driver of retention. Customers who encounter a consistent brand experience across digital, email, social, and physical contexts develop stronger brand relationships than customers who experience inconsistency. The trust that coherence builds reduces the need for discounts to drive repeat purchase. **How to apply this:** Map every customer touchpoint and ask: does this express the same brand beliefs as every other touchpoint? If the website says premium materials but the packaging uses generic cardboard, the coherence is broken. If the Instagram content is aspirational but the email subject lines are promotional and discount-first, the coherence is broken. Each break in coherence erodes the brand relationship you are trying to build. *** ## The Transferable Framework Buck Mason's position was not built from a single strategic decision. It was built from a set of consistent choices, made over time, that reinforce the same underlying belief: substance over style, material over marketing, full-price integrity over promotional volume. The brands that most successfully apply these frameworks are the ones that identify the equivalent of "fabric-first" in their own category — the dimension of quality or substance that most other brands treat as secondary — and build the entire product, pricing, and content architecture around it. Click **Open in Claude** above and share your brand's URL. Claude will analyze your site against these five Buck Mason frameworks and identify where your information architecture, pricing philosophy, and content strategy are and are not coherent. # Huckberry Brand Teardown: Editorial Commerce and the Curated Discovery Model Source: https://intelligence.madebydas.com/teardowns/huckberry A research brief on Huckberry's DTC strategy — editorial-first product discovery, the 'store' as a magazine, community-driven content, and the curation model that replaces search with editorial authority. Part of the DAS brand teardown series. Expanded analysis coming soon. # Huckberry Brand Teardown *Research brief — expanded analysis coming soon.* ## The Core Thesis Huckberry built a \$100M+ DTC business without being a brand in the traditional sense. They are an editorial platform that sells products. The distinction matters. A traditional DTC brand manufactures or sources products and uses content to drive awareness and conversion. Huckberry inverts this: they create content — stories, features, profiles, adventure narratives — and the products exist inside that content as tools for a way of life the editorial voice has already made aspirational. The result is a buying experience that does not feel like shopping. Customers browsing Huckberry are reading. The products surface as recommendations inside a narrative, not as items in a catalog. Conversion happens almost incidentally, which is why it happens at higher rates than conventional catalog browsing. *** ## What Huckberry Built **The store as a magazine.** Huckberry's product pages read like editorial. A waxed canvas jacket is not described by material weight and construction — it is described in the context of a winter camping trip, or a morning walk through a cold city, or the first hike where you realized you needed something that could handle weather. The product exists inside a story that the target customer wants to be in. **Curation as authority.** Huckberry does not carry every product in a category. They carry the product they believe is the best in the category, sourced from makers whose story aligns with the editorial voice. This curation is the product. The selection signal tells the customer: we've already done the research. You don't have to. **Community as content engine.** Huckberry's community — the men who read the editorial, buy the gear, and identify with the lifestyle — generates content through submission, tagging, and feature opportunities. The brand provides the platform and the editorial standard. The community provides authentic documentation of the brand's products in use. **Email as the primary channel.** Huckberry's email program is widely cited in DTC circles as one of the highest-performing in the industry — not because of sophisticated automation, but because the email itself is worth reading. The editorial voice carries into the inbox. Subscribers read the emails the way they would read a magazine — not because they are waiting for a promotion, but because the content is genuinely interesting. *** ## The Frameworks *(Detailed analysis and transferable frameworks pending expanded teardown. The three frameworks below are drawn from the existing research brief.)* ### Framework 1: Curation Reduces Decision Fatigue and Increases Conversion When a customer trusts the curator's judgment, the decision is already made before they reach the product page. Huckberry's curation standard — one product, from one maker, with one story — removes the paralysis of choice that most catalog shopping creates. The customer is not choosing between twelve similar jackets. They are deciding whether they want the one jacket Huckberry has decided is worth carrying. **For DTC brands:** Fewer SKUs presented with editorial authority converts better than broader catalogs presented with neutral copy. If every product you carry is described as if you'd stake your reputation on it, the customer's trust in your selection becomes the conversion driver. ### Framework 2: Editorial Commerce Attracts a Different Buyer Huckberry's customer is not a deal-hunter. They are buying into a lifestyle — the gear is proof of membership in a particular kind of life that the editorial voice has made vivid. This customer has higher LTV and lower price sensitivity than the average DTC buyer, because the transaction is not about the price point — it is about the identity. **For DTC brands:** If your content is purely promotional, you are attracting buyers. If your content is editorial, you are attracting members. Members have higher LTV. The difference is whether the content is worth reading when you are not planning to buy something. ### Framework 3: The Email Program as Magazine Subscription Huckberry's email program works because the email is the product, not the vehicle for the product. Subscribers receive stories, gear recommendations embedded in narratives, and access to limited drops — not promotional emails with discount codes and urgency timers. **For DTC brands:** The test for your email program is simple. Would your subscribers read the email even if they were not planning to buy anything this month? If the answer is no, you are running a promotional email program, not an editorial one. Promotional email programs generate revenue during the campaign and churn afterward. Editorial email programs build relationships that generate revenue over years. *** ## Coming Soon The full Huckberry teardown will include: * An analysis of their editorial archive and how it compounds over time * The product selection process and the curation framework * Email program structure and cadence analysis * Community content mechanics and how user-generated content feeds the editorial voice * The brand's approach to influencer and creator relationships Click **Open in Claude** above and share your brand's URL and email program. Claude will analyze whether your content strategy, email program, and product presentation follow the editorial commerce model or the promotional model — and what the conversion and LTV implications are. # Rose Los Angeles Brand Teardown: Collaboration-as-Acquisition and Capped Membership Source: https://intelligence.madebydas.com/teardowns/rose-los-angeles A DTC brand teardown of Rose Los Angeles — $900/year Rose Club capped at 2,000 members, chef collaborations with Enrique Olvera and Dominique Crenn, engineered waitlists as owned audience infrastructure. Three transferable frameworks for DTC brands. Part of the DAS brand teardown series. # Rose Los Angeles Brand Teardown ## A \$40 Edible Empire Built Without Paid Ads Rose Los Angeles cannot run ads on Meta, Google, or TikTok. Regulatory restrictions ban cannabis brands from every major paid acquisition channel. Yet Rose has earned coverage in Vanity Fair, Vogue, Bon Appétit, the Wall Street Journal, and Fast Company (named #10 Most Innovative Wellness Company in 2022) — and sells Turkish delight-style cannabis edibles at \$40 per box, roughly 2–3× the industry average. They did it by turning three constraints into compounding advantages that any DTC brand can replicate. This is the playbook for CMOs who can run paid ads but are building on rented land. *** ## The Brand Rose Los Angeles was co-founded in 2017 by Nathan Cozzolino and Scott Barry. Cozzolino, a former actor diagnosed with stage 3 cancer in his late 20s, started growing cannabis to manage chemotherapy symptoms. Barry, a CalArts-educated graphic designer, co-founded PlantPaper (tree-free toilet paper) and previously worked in the LA food scene. The format — Turkish delights — came from C.S. Lewis. The product is a potato starch-based, vegan, gluten-free confection infused with single-strain flower rosin, a solventless extract mechanically pressed from whole cannabis flower. Rose claims to be the first edible company in the recreational market to exclusively use flower rosin, pressed in-house at their facility in San Francisco's Mission District. What makes Rose relevant to DTC brands outside cannabis is not the product — it is the model. Rose was forced to build what most DTC brands say they want but never prioritize: a brand that generates its own demand without renting distribution from Meta and Google. *** ## Broken Rule 1: The Product Is the Content Calendar Most DTC brands run two parallel tracks — product development and content marketing — staffed by different teams with different goals. Rose collapsed them into one function. Every new product is a content event. The mechanism: Rose partners with world-class chefs to co-develop limited-edition seasonal products. Enrique Olvera (Pujol, Cosme) has done multiple editions. Virgilio Martinez (Central — ranked #1 restaurant in the world) co-created "Magical Grapes," now in its fourth consecutive season. Dominique Crenn (Atelier Crenn — first female chef in the US to earn three Michelin stars) developed an apple-passionfruit edition. David Zilber, former head of Noma's fermentation lab, made a kimchi-pear version. The list extends beyond the kitchen: Brad Leone (YouTube star), Andy Baraghani (Bon Appétit alum), Kate Berlant, Susan Sarandon, the band MUNA, comedian Hannah Einbinder, and film studio A24 (a sleep-oriented "Dream Scenario" edition tied to the Nicolas Cage film). **Why this matters for non-cannabis DTC brands:** Each collaboration generates a minimum of four marketing outputs simultaneously — a press angle, an email event, social content from the collaborator's audience, and a scarcity trigger. Rose does not need a content team brainstorming "what to post this week." The product roadmap is the content calendar. **The transferable framework:** Treat 3–4 product launches per year as co-created editorial events with culturally relevant partners whose audiences overlap yours. The collaboration pipeline becomes a predictable content and acquisition engine that compounds as the network grows. *** ## Broken Rule 2: Shrink Distribution to Grow the Brand In January 2024, co-founder Nathan Cozzolino made a decision most DTC operators would consider suicidal: Rose stopped producing wholesale to sell into retail locations. His reasoning was blunt: "There just isn't the margin to support continuing to use the quality of ingredients that we choose to use." Rose walked away from dispensary shelves and retail presence to go DTC-only, shipping nationwide using hemp-derived THC under the 2018 Farm Bill. This is the opposite of the standard CPG playbook. The retention architecture has three tiers: **Subscribe & Save** — 15% off on core products (dropping the \$40 box to \$34). **Rose Club** — \$900/year, capped at 2,000 members, opening registration once annually on December 11. Members receive quarterly care packages (6 products per quarter, minimum \$1,020 retail value), members-only gifts worth \$240, a 20% discount, early access to seasonal drops, and — critically — a personal onboarding phone call, a dedicated phone line, and a direct email answered only by the Rose team. Rose explicitly positions this as the anti-subscription: "Rose Club is not a product, it is a sustained relationship of reciprocal care." **Structural scarcity** — Products sell out routinely. The Brad Leone Chokeberry, Andy Baraghani Sumac Meyer Lemon, and David Zilber editions are either sold out or on waitlists. This scarcity is not artificial — it is a natural consequence of using peak-season ingredients from named farms. When the harvest is gone, the product is gone. **The counterintuitive insight:** Rose's scarcity turns a supply chain constraint into a brand-building asset. Customer reviews confirm the behavior this creates: "Some of their collabs sell out VERY quickly. If you find that you like these delights, be ready to jump when you see an email about a limited edition flavor." At theoretical full Rose Club capacity — 2,000 members at \$900/year — that is \$1.8M in annual recurring revenue before any additional purchases. *** ## Broken Rule 3: Compete in the Category Next Door Rose does not appear in cannabis trade publications. It appears in Vogue, Vanity Fair, Bon Appétit, the Wall Street Journal, New York Times Style, the Los Angeles Times, W Magazine, and New York Magazine's The Strategist. This is not accidental. Rose's competitive set is not other cannabis brands. It is artisanal food. Their product copy reads like a tasting menu. They name 20+ sourcing farms across their catalog — Blossom Bluff Orchards in Parlier, Dirty Girl Farm in Santa Cruz, Rincon Tropics in Carpinteria, Lagier Ranches in Escalon for rare Seville oranges. A Thingtesting reviewer captured the positioning: "One of the best brands in cannabis. Their approach is that of a Michelin-starred restaurant hidden in the drugstore candy aisle." Another: "Wish you made Delights without the cannabis so I could eat these all the time." When customers want the product without the active ingredient, the brand has transcended the category. **The framework:** Rose chose to be evaluated by food and culture editors, not cannabis reviewers. That decision determined everything downstream — sourcing standards, collaborator selection, brand voice, and which audiences discovered them. The transferable lesson: identify the adjacent category where your brand could credibly compete, then build the product and content standards that belong in that higher-status category. *** ## Three Transferable Frameworks ### Framework 1: Collaboration-as-Acquisition Rose turns product development into a customer acquisition channel. Each chef or cultural collaborator brings their audience, generates earned media, and creates a time-limited reason to buy. For a DTC brand at \$15M–\$100M: treat 3–4 product launches per year as co-created editorial events with partners whose audiences overlap yours. The collaboration pipeline becomes a predictable content and acquisition engine that does not require paid media to drive reach. ### Framework 2: Engineered Waitlists as Owned Audience Infrastructure Rose's sold-out products are not failures of demand planning. They are email capture mechanisms. Every sold-out SKU has a "Notify Me" button. Every seasonal drop trains customers to watch their inbox. The behavioral loop: anticipation → email notification → purchase → sell-out → waitlist for next drop. This creates habitual email engagement that Klaviyo flows alone cannot generate. For any DTC brand: if nothing on the site is ever sold out, you are leaving owned audience growth on the table. Structural scarcity — through ingredient sourcing, production capacity, or membership caps — is an earned media and retention driver, not a supply chain failure. ### Framework 3: The Concierge Membership Model Rose Club at \$900/year, capped at 2,000 members, with a personal onboarding phone call and a direct email answered by a human. At full capacity, that is \$1.8M in annual recurring revenue from 2,000 members, before the 20% discount on additional purchases drives further revenue. The model works because the relationship is genuinely personal, not performatively personalized. Rose's own language: "In an era where online commerce simulates intimacy through algorithms and automated subscriptions, we choose a different path." For DTC brands building loyalty programs: the counterintuitive move is to cap membership, charge significantly more, and deliver concierge-level service rather than scaling a points program to millions of low-engagement members. *** ## The Proof of Concept Rose's advertising restrictions forced them to build exactly what the DAS methodology predicts compounds: an owned audience, an editorial brand, and a retention model that does not depend on re-acquiring the same customer through paid channels every 90 days. The brands at \$15M–\$100M that will build the most durable positions over the next five years are the ones that act as if Meta could shut off their ad accounts tomorrow. Rose proves this works — not in theory, but at \$40 per box, with a waitlist to prove it. Click **Open in Claude** above and share your brand's URL. Claude will analyze your site against these three Rose LA frameworks and identify whether your content strategy, membership model, and scarcity architecture are working for or against your owned audience growth.