Updated: Oct 27
Klaviyo Predictive Analytics: Turn CLV Into Useful Decisions
By Trayan Hristov · Thrivelia · Reviewed October 2026

Klaviyo predictive analytics can help an established store estimate future customer spending and choose useful audiences or timing tests. Use those estimates to guide a decision, then check what customers actually do. A predicted value is neither a promised purchase nor a measure of customer profit.
The useful question is specific: should this group receive a different product introduction, service experience or reminder? Answer that question before creating another high-value segment.
Check eligibility before planning around the fields
Klaviyo's dedicated predictive analytics guide sets these requirements for profile CLV predictions:
At least 500 purchasing customers with nonzero orders that were not cancelled or refunded.
An ecommerce integration or API-supplied order data.
At least 180 days of order history, with orders within the last 30 days.
Some customers who have placed three or more orders.
These are purchasing customers, not a count of subscribers or orders. The three-order condition applies to the account, rather than every recipient. Inspect individual profiles too: a missing prediction is unknown, rather than evidence of no future value.
Make the purchase history trustworthy
Before using a forecast, trace representative orders from the store into Klaviyo. Include a full-price order, a discounted basket, a partial refund, a cancellation and a repeat customer. Check dates, values, currency context and whether one person has been split across profiles.
Review the account's mapped metrics with the person responsible for reporting. Those mappings tell predictive models which events represent the relevant commercial activity. A custom integration or mixed online-and-retail business needs particular care about which order sources are included.
Record the current definitions before making changes. Changing a mapping to improve one model can affect other applicable reports. Investigate duplicate events and missing history at the source instead of adjusting a forecast until it looks plausible.
Separate past spending from the prediction period
Klaviyo's CLV definitions distinguish three values:
Historic CLV describes recorded past order value under the account’s configuration, including the relevant adjustments for refunds and returns.
Predicted CLV estimates spending during the configured future period.
Total CLV adds historic and predicted CLV. It is not all spending expected from today onwards.
Keep contribution separate. Product cost, discounts, fulfilment and service costs affect what a customer relationship is worth commercially. A spending forecast alone cannot tell you how much to pay to acquire or retain that customer.
The default prediction period is 365 days. Custom CLV settings can change that period, and the change affects segments, flows, campaigns and forms using predicted CLV. Record the horizon with every threshold and export; a value of 200 over 90 days expresses a different forecast from 200 over a year.
The custom CLV dashboard and associated capabilities have Personalization or Advanced KDP entitlement requirements. They are separate from the standard marketing application. Confirm account access before planning around custom settings; do not assume every profile prediction requires the same add-on.
Match the field to a customer decision
Signal | Possible treatment | Check before acting |
|---|---|---|
High historic CLV | Recognise an established customer relationship. | Past spending does not establish future intent or profit. |
Higher predicted CLV | Test a relevant product introduction. | Check the forecast period, stock and customer context. |
Expected next-order date | Test a useful reminder around likely timing. | Check whether the purchased product has a known cycle. |
Higher churn prediction | Investigate a barrier to returning. | Purchase risk does not establish dissatisfaction or email permission. |
Klaviyo models churn risk from order behaviour. It does not tell you that the person is ignoring email, has withdrawn consent or needs a larger discount. Keep marketing eligibility and the actual customer problem as separate decisions.
When building an opted-in email audience, combine the predictive condition with appropriate eligibility and subscribed status, as Klaviyo's CLV segmentation guide illustrates. Add relevant purchase and service exclusions. A valuable prediction does not override a customer's channel permissions.
Our segmentation guide covers the wider treatment brief: what changes for this customer, why it should help and when it stops being relevant.
Give expected next-order dates the right job
The next-order explanation combines the customer's order behaviour with account-wide patterns. Klaviyo's Catalog Insights comparison makes an important distinction: expected next-order date does not consider which products were bought.
That matters for a fictional outdoor retailer selling boots, waterproofing treatments and seasonal accessories. A predicted order date does not establish that a customer needs new boots. The useful next message might be care guidance or a compatible accessory.
Where product consumption gives you a known reorder cycle, use a product-specific replenishment journey. Check quantity, subscription status and subsequent purchases. Keep the predicted date as another hypothesis rather than replacing better product evidence.
Validate the forecast on a dated customer snapshot
A segment changing every day is useful for activation, but makes a poor retrospective record. Preserve what the model predicted before the later orders happened. Otherwise, an updated prediction can accidentally use the behaviour you are trying to evaluate.
Choose a snapshot date and record the included customers, model horizon, mapped metrics and forecast values.
Create mutually exclusive forecast bands, such as lower, middle and higher predicted spending. Keep profiles with missing values separate.
After the full forecast period, calculate observed qualifying spending for those same customers using matching definitions.
Compare predicted and observed totals by band, plus customer count and average spend. Inspect whether the bands meaningfully separate later behaviour.
Investigate misses by acquisition period, first product, returns, stock changes and promotions before changing the customer treatment.
The CLV segment export guidance explains how to export predictive fields for group analysis. Use the appropriate export permissions and retain a dated definition. A short pilot cannot validate a full-year revenue forecast: label partial follow-up clearly and avoid comparing different horizons as though they match.
Klaviyo describes predictions as more useful across groups than exact for individuals. Your validation should show the size and direction of uncertainty that matters to the business, rather than labelling one customer’s unexpected order a model failure.
Test the treatment separately from forecast quality
Return to the fictional outdoor retailer. Suppose a predicted-spending band identifies first-time buyers for a compatible-accessory message. Freeze the eligibility rule, exclude customers for whom the product or timing is unsuitable, and define the order and contribution outcomes before launch.
Where volume permits, randomly split that eligible audience between the proposed treatment and a suitable control while keeping the rest of the programme comparable. Measure the difference after a fair observation window, including discount cost and unsubscribes.
A higher-value group spending more than a lower-value group can support the model’s usefulness. It does not establish that the message caused the spending. The treatment comparison answers that second question.
Keep working when the account is not ready
Start with observed purchase stages, first products and time since purchase. Use equal-age first-to-second purchase cohorts to understand actual customer progression. Improve event quality and history while developing relevant journeys; predictions are not a prerequisite for helpful lifecycle marketing.
Frequently asked questions
Is predicted CLV the same as customer profit?
No. It estimates customer spending over the prediction period. Assess costs and contribution separately. Total CLV also includes historic value, so it should not be treated as future revenue alone.
Can I use churn risk to suppress email subscribers?
It is a purchase-behaviour estimate, not a marketing-permission or email-engagement decision. Review the actual reason for suppression and the person’s channel status independently. High predicted churn alone is insufficient.
Should expected next-order date replace replenishment flows?
Use known product cycles when they provide better timing evidence. The expected date does not consider purchased products. Test its usefulness for the specific audience and coordinate it with existing journeys.
For help connecting predictive fields to practical journeys and measurement, explore our Klaviyo agency support and retention marketing services. Start with one customer decision and a way to assess it.

.png)
Comments