25 Klaviyo Micro-Segments Generated by AI — Segment Your List Without a Data Team

Two years ago I watched a DTC skincare brand burn through $14,000 in ad spend before they realized their "active customers" segment was 22% of the list, but 78% of those hadn't actually purchased in over a year. The segment definition looked right. The reality was a graveyard of one-time holiday shoppers they were hitting with daily broadcast emails. Unsubscribe rate was 0.9% per send. The list was quietly dying.

Klaviyo's default segments are like that. They're serviceable. They're also blunt. "Active customers" tells you almost nothing useful — you need to know what kind of active, what they did, what they're likely to do next, and what message will land.

The marketing directors I work with don't have data engineers. They have a Klaviyo login, a CSV export, and ChatGPT. That's enough.

Here's the workflow: you dump 90 days of Klaviyo events into a spreadsheet (or paste into Claude), give the model a structured prompt, and walk away with 25 micro-segments that you can build inside Klaviyo in an afternoon. The segments below are the actual 25 I generated for the skincare brand above — minus their names, plus the reasoning.

The AI prompt that produces all 25

Before I give you the segments, the prompt. This is the part you actually save. Drop your Klaviyo event export (or a synthetic list with the same columns) into Claude/ChatGPT and use this:

You are an ecommerce lifecycle strategist. I'm giving you 90 days of Klaviyo events for a [category] brand with [X] active subscribers. The columns include email/SMS events, Placed Order, Added to Cart, Viewed Product, Custom events, plus profile properties (tags, predicted CLV [Customer Lifetime Value, 客户终身价值], predicted gender, source, location).

Generate 25 micro-segments grouped into 5 categories: (1) Lifecycle stage, (2) Behavior signal, (3) Predictive AI, (4) Channel & consent, (5) Strategic niche. For each segment give: name, definition in plain English, the Klaviyo segment builder recipe (conditions + AND/OR logic), the campaign angle that works for that segment, and one watch-out.

The output is a document you can paste straight into Notion. Below is the rendered version of what came out — and the 5 cluster principles I now use on every Klaviyo account I touch.

Cluster 1: Lifecycle stage (5 segments)

This is where the biggest revenue leaks hide. "Customer" is not a segment. It's a state someone is passing through.

1. Day 0–6 newsletter readers, no purchase yet. Warm but unconvinced. Klaviyo's default welcome series treats them identically to a recent buyer. Don't. Send the social proof and objection-handling — not the discount.

Recipe: What someone has done is Subscribed to list exactly in the last 6 days AND Placed Order zero times.

2. First-purchase 7–21 day window. They're 3x more likely to buy again in this window than at any point after. Cross-sell the natural next product, not a generic "thank you."

Recipe: Placed Order at least once AND Placed Order between 7 and 21 days ago AND Number of orders is less than 3.

3. Second-order lapsed (bought once, no second buy in 60–90 days). This is the silent churn segment. Most brands have 4x more of these than they realize.

Recipe: Number of orders is exactly 1 AND Placed Order between 60 and 90 days ago.

4. Loyalty peak (3+ orders, recent activity). The top 8% of your list, often 35-40% of revenue. Don't send them discounts. Send them access.

Recipe: Number of orders is at least 3 AND Placed Order in the last 45 days.

5. Cold revival candidates (90+ days, no opens, still on list). Klaviyo's default is to let them rot. With 90 days of zero engagement, a single re-engagement email is a 50/50 — but a 3-email re-engagement flow over 14 days recovers 4-7% of them, which on a 100k list is real money.

Recipe: Has not opened email in the last 90 days AND Has not unsubscribed ever AND Number of orders is at least 1.

The wrong way to do this: send the same "we miss you" email to all 5 of these. The right way: each one gets a different message because they're in genuinely different states.

Cluster 2: Behavior signal (5 segments)

Behavioral micro-segments outperform demographic ones by 3-5x on revenue per send. This is the cluster most teams under-build.

6. Browse-abandoners (24h). They viewed a product page but never added to cart. Soft intent. The best-performing subject line for this group is rarely a discount — it's the object of curiosity they were already looking at.

Recipe: Viewed Product in the last 24 hours AND Added to Cart zero times AND Placed Order zero times.

7. Cart-abandoners, 1-hour window. The peak conversion moment. Most Klaviyo templates start the abandon flow at 4 hours. You lose 30% of the possible conversions that way.

Recipe: Added to Cart in the last 1 hour AND Placed Order zero times since adding to cart.

8. Wishlist adders, no purchase. High intent, high consideration. Different emotional driver from cart abandoners — they want validation, not urgency.

Recipe: Added to Wishlist in the last 7 days AND Placed Order zero times AND Has not been in segment "Recent buyers".

9. Pricing-page viewers. Self-qualified bottom-of-funnel. If you have a pricing page (for subscription or tiered products), this is your hottest non-buyer segment.

Recipe: Viewed Page where Page name contains "pricing" in the last 14 days AND Number of orders is 0.

10. Repeat site visitor, no purchase (3+ sessions). They keep coming back. They keep not buying. Usually a friction or trust problem the analytics won't show you — and exactly the people you should survey.

Recipe: Number of sessions is at least 3 in the last 30 days AND Placed Order zero times.

Watch-out for this whole cluster: make sure your Klaviyo event tracking actually fires. If Viewed Product isn't being captured (it's a separate Klaviyo snippet from Placed Order), segments 6 and 9 are empty. Audit the Klaviyo JS (JavaScript) snippet first.

Cluster 3: Predictive AI (5 segments)

Klaviyo ships with five predictive features most accounts never use. This is the cluster where AI-as-segment-builder pays for itself fastest — because the model is doing the work, you just have to read the output.

11. Predicted CLV top decile (top 10% by predicted lifetime value). Send this group the brand-building emails: early access, founder notes, behind-the-scenes. The worst thing you can do is give them a 15% off coupon — they were going to pay full price.

Recipe: Predicted CLV is in the top 10%.

12. Predicted next order within 7 days. Klaviyo's predictive analytics gives you a per-profile "next order date." When that window is now, you have a high-confidence send moment.

Recipe: Predicted next order date is within the next 7 days AND Placed Order zero times in the last 7 days.

13. Predicted gender split (for gendered catalogs). Klaviyo's predicted gender property is wrong about 8-12% of the time, but for a brand with male/female SKUs (apparel, beauty, wellness) it splits the list into two mail streams that each outperform a unisex send.

Recipe: Predicted gender is "female" (build a parallel male version).

14. Engagement score tier 1 (the engaged top quartile). Klaviyo's email engagement score (0-100) is one of the most underweighted signals. The top quartile consistently drives 60%+ of clicks.

Recipe: Email engagement score is at least 75.

15. Predicted to churn (the Klaviyo "Likely to Churn" segment). Built into Klaviyo. Most accounts have it but don't message it differently. Don't broadcast to it. Send it to a 3-email save flow with a real offer.

Recipe: Predicted gender is "Likely to Churn" (Klaviyo's own property) AND Number of orders is at least 2.

The mistake to avoid: building 15+ flow variations off these predictive segments. Pick 2-3 and let the data accumulate for 60 days. Predictive properties need volume to stabilize.

Cluster 4: Channel & consent (5 segments)

Channel-specific segments prevent the worst email-marketing failure mode: sending a 1,200-word story to someone who gave you their phone number.

16. SMS consented AND email engaged. The omnichannel hot list. Best ROI for time-sensitive offers.

Recipe: Consented to receive SMS is true AND Email engagement score is at least 50.

17. SMS only, no email engagement. They want texting. Respect that. Sending them email is the path to "STOP."

Recipe: Consented to receive SMS is true AND Number of email opens in the last 60 days is 0.

18. Push notification opted-in. Often the highest-revenue segment on Shopify mobile apps. Most brands don't email it separately.

Recipe: Consented to receive push notifications is true (Shopify integration) AND App sessions in the last 30 days is at least 3.

19. International subscribers (non-home country). Currency mismatch kills open rates. So does shipping copy that doesn't apply. This segment is its own send calendar.

Recipe: Country is not [your shipping country] AND Email engagement score is at least 30.

20. Unsubscribed from a flow but not the brand. They hit unsubscribe on the abandoned cart sequence. The brand relationship isn't gone. Most teams exclude them from everything — which is wrong.

Recipe: Unsubscribed from is "Abandoned Cart Reminder" AND Has not unsubscribed from all marketing.

Cluster 5: Strategic niche (5 segments)

These are the segments your marketing director will think of last. They're the ones with the highest per-recipient revenue.

21. Discount-only buyers (used a code in last 60 days). Price-sensitive cohort. Send full-price launches to them last, or not at all — and you can spot margin erosion before it shows up in P&L (Profit & Loss).

Recipe: Used a discount code in the last 60 days.

22. Full-price buyers (no code in last 90 days). The cohort you can run early-access drops to. They pay full price, so your margin dollars are real.

Recipe: Has not used a discount code in the last 90 days AND Placed Order in the last 90 days.

23. Recent reviewers (left a review in last 30 days). The single most engaged post-purchase cohort. They're already telling their friends. Reward the behavior — don't just ask for more.

Recipe: Left a review in the last 30 days (custom event) AND Number of orders is at least 1.

24. VIP / wholesale tags. Many Klaviyo accounts have wholesale buyers mixed into the DTC list. The wholesale buyer opening your "30% off sitewide" email is a margin loss.

Recipe: Tag contains "wholesale" OR Tag contains "VIP".

25. Replenishment cycle window. For consumables (skincare, supplements, pet food), predict the reorder window from the customer's last order. Different cohort, different send cadence.

Recipe: Placed Order with SKU in collection [consumable collection] in the last [cycle length] days AND Placed Order zero times in the last [cycle length - 7] days.

The watch-outs nobody tells you

Three things will go wrong with this list. They go wrong on every Klaviyo account I work on. Plan for them.

Event tracking is incomplete. "Viewed Product" and "Added to Wishlist" don't fire by default — they need the Klaviyo JS snippet installed on the relevant pages, plus a custom event setup. Segments 6, 8, and 9 will be empty until that's done. Audit first, build second.

Predictive properties need 90+ days of order data. If your store is newer, Klaviyo's predicted CLV, predicted gender, and predicted next order date properties return "unknown" for most profiles. Wait the 90 days or skip the predictive cluster.

Segment overlap is a feature, not a bug. A subscriber can be in three of these at once. That's correct. The mistake is trying to write unique copy for every overlap combination — Klaviyo's flow logic handles that with priority and exclusion rules. Don't try to build it in segments.

The part that isn't about segments

The 25 segments above are useful. The bigger shift is the workflow. Two years ago, building a Klaviyo segment required a marketing ops person to interpret the data, a strategist to define the angle, and a copywriter to test the message. That sequence was a week.

The AI workflow is: export 90 days of events, run the prompt, get 25 segments with definitions and angles, build them in Klaviyo in an afternoon, write the email against the segment you trust most. The bottleneck moves from "what should I segment on" to "what should I say to them." That second question is the only one AI doesn't fully answer yet — and it's the only one that ever really mattered.

The list above isn't a static document. Run the prompt again in 60 days, on the new 90 days of events. The segments evolve with your list, your season, your catalog. The prompt is the asset, not the output.