How to Segment Klaviyo Customers Beyond “Engaged in the Last 90 Days”

“Engaged in the last 90 days” can help determine who is eligible to receive a campaign. It does not tell you what to say to them.

Inside that audience may be a subscriber who has never purchased, a customer who ordered yesterday, a repeat buyer exploring another category and someone waiting for a refund. They do not have the same relationship with the brand or the same reason to buy.

Useful segmentation connects what you know about a customer to a decision you will make differently. If two segments receive the same products, message, timing and offer, the extra segmentation may not be doing much.

Start with the decision

Before building a segment, complete this sentence: “Because this customer has done or told us X, we will change Y.”

For example, a customer who bought a sample kit needs help selecting a full-size product. A customer who has just bought the full-size product needs onboarding. A customer who returned it may need service or a different recommendation.

The segment is the technical expression of that decision. The strategy is understanding why the experience should change.

Klaviyo’s custom profile property documentation explains how properties can support segments, flow filters and dynamic content. The useful question is which property or event actually changes the next experience.

1. Separate purchase stages

Start with non-buyers, first-time buyers and repeat buyers. Within those groups, use purchase recency and the expected product journey.

Non-buyers may need confidence, product selection and evidence. First-time customers may need adoption and reassurance. Repeat buyers may respond to complementary categories, new releases or a better service experience.

Avoid assuming that every repeat buyer is a VIP or every non-buyer needs a discount. The segment should guide a hypothesis, not replace judgement.

Define entry and exit conditions. Once a subscriber places an order, they should not keep receiving first-purchase messages that conflict with that order.

2. Use the first product to guide the next step

The first purchase often tells you more than broad demographic assumptions.

For apparel, distinguish someone who bought a staple T-shirt from someone who bought occasionwear. For beauty, separate discovery-kit purchasers from customers who already selected their full-size product. For pet products, consider the product type and any voluntarily supplied pet needs.

Then investigate what customers in each cohort actually buy next. Do not assume your preferred cross-sell is the natural next purchase.

Our fashion and apparel guide and beauty and skincare guide show how the commercial journey changes by category.

3. Build timing around the buying cycle

A 60-day lapse means different things for coffee, footwear and furniture. Use observed time to the next purchase, product consumption and seasonality to define useful windows.

A replenishment hypothesis should account for pack size and subscription status where available. A durable-product customer may need care, accessories or a gift recommendation instead of a reminder to buy the same item again.

Treat the expected interval as a range rather than an exact prediction. Test timing and keep a control where volume supports it. Sending on the estimated depletion date does not prove that the estimate is correct.

4. Combine interest with ownership

Browsing a product is more useful when combined with purchase history.

A customer viewing a product they already own may be looking for care instructions, a replacement or a gift. A non-buyer browsing the same item may be checking fit, value or suitability.

Use recent browsing and clicks alongside ownership, stock and availability. Check whether the product is appropriate in the customer’s market. Ensure that an actual purchase ends the relevant recovery treatment.

This is how segmentation moves from “interested in category” to a useful next action.

5. Respect current service context

A customer with an unresolved delivery problem is not necessarily ready for another promotional push. Where the data and permission allow, use reliable service signals to pause or adapt relevant campaigns.

Define the rule narrowly. A resolved support ticket should not suppress someone forever. Likewise, an unknown service status should not be treated as confirmed dissatisfaction.

Document which system is authoritative, how quickly updates arrive and who owns errors. A segment is only as dependable as the data feeding it.

6. Collect preferences progressively

Ask for information when it improves the immediate experience. A preference centre, product finder or post-purchase question can reveal useful context without turning every sign-up into a questionnaire.

Explain how the answer will be used. Keep unknown as a valid state, provide a sensible fallback and allow preferences to change.

The principle is developed in our Progressive Personalisation guide: the relationship should become more useful as the customer gives you more evidence, rather than asking for everything at the start.

A practical segment specification

For every important segment, document the business question, source data, inclusion rule, exclusion rule, exit condition, treatment and success metric.

A hypothetical starter-kit segment might include customers whose kit was delivered 7 to 21 days ago and who have not bought full size. Exclude returns and relevant unresolved service cases. Exit immediately after a full-size purchase.

The treatment is product-selection guidance. The primary outcome is full-size purchasing within a defined window, with margin, returns and unsubscribes as guardrails. The dates are a test hypothesis, not a universal benchmark.

QA the people, not just the rule

Sample real profiles and check whether their membership makes sense. Include people who should qualify, people who should not and edge cases with missing information.

Look for accidental AND/OR combinations, overlapping offers, outdated properties and exclusions that remove nearly everyone. Estimate the audience before launch and investigate surprising changes.

Keep an owner and a dated change log. A technically valid rule can still express the wrong commercial decision.

Measure whether the distinction mattered

Compare the tailored treatment with an appropriate control within the same eligible population. Comparing one naturally high-value segment with a low-value segment does not demonstrate that personalisation worked.

Review completed behaviour, contribution and later purchasing alongside channel metrics. In perfectwhitetee’s public case, customer analysis and journey development sit alongside reported channel outcomes; the case should not be read as a controlled test of one segmentation rule.

If your audience strategy stops at recent engagement, book a consultation. We can identify which customer distinctions are commercially useful and turn them into journeys your team can maintain.

Continue with our guide to trustworthy engagement signals and Klaviyo strategy and implementation.