​

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. 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. 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.

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

→ 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

---

​

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

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

​

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.