Klaviyo Predictive Analytics: Wiring the Fields Nothing Reads
Klaviyo already predicts which customers will churn and what they will spend next year. In most accounts, no segment, flow, or campaign reads those fields. The play is the wiring.
The predictions are on. Nothing reads them.
Open a qualifying account and the fields are there.
What Klaviyo predicts, and predicts well
Every qualifying account carries three predictive fields on the profile itself, per Klaviyo's docs: predicted customer lifetime value, churn risk, and the expected date of the next order, all retrained weekly. The table below carries the specifics; every row is attributed.
Per Klaviyo's predictive analytics documentation, checked July 8, 2026. Details move; the docs are the source of truth.
Three wires, native conditions.
-
Build the two segments
Cross churn risk with predicted CLV, in native segment conditions. Rising risk on a high-CLV profile is the save list: your best customers, flagged before recency ever moves. Rising risk on a low-CLV profile is the let-go-cheaply list: one inexpensive touch, then quiet. Two segments, built in minutes, and the account starts sorting who deserves effort from who does not.
-
Trigger on the transitions
Trigger the flow on the crossing itself, the moment a profile moves between bands. Low to Medium fires the cheap automated touch: a reminder, a restock note, the kind of send that protects margin. Medium to High on a high-CLV customer fires the concierge save, the send a human would write. The treatments themselves, discounts held for last, follow the winback ladder.
-
Route timing through the date
The expected date of the next order is a clock, and clocks are for scheduling. Anchor the refill nudge to that date and the timing follows each customer's own cycle. Then watch the other direction: a customer quietly past their expected date is the earliest miss signal the account has. Both jobs, the nudge and the miss, run inside the replenishment flow.
Twenty-five years of model behind the fields
These fields productize the buy-till-you-die family of models: Pareto/NBD and its more tractable successor BG/NBD, developed by Fader and Hardie. The models estimate each customer's latent "still alive" probability from recency and frequency, with a Gamma-Gamma extension for monetary value, linking RFM directly to lifetime value. The Wharton paper and the BG/NBD explainer are linked in Sources.
The same math this series teaches by hand: the scores, the buckets, the migration view. Treat the predictions as the automated cousin; Fader and Hardie call RFM nearly a sufficient statistic for these models.
Where the fields stop, per Klaviyo's own docs
Per Klaviyo's documentation, the churn bands are coarse by design: Low, Medium, High, and the banding is the whole resolution. The expected next-order date turns generic for one-time buyers, estimated from your whole customer base. And every field is blind to your margins. All of that is design scope. The fix is judgment layered on top of the fields, the same posture we took in our Composer review.
Two reads, on your own data.
Intervened vs. untouched
Compare churn among customers the wiring touched against those it left alone, read within the same CLV band on your own data. No published benchmark exists for this page to quote; the read is the gap between your two curves.
The save-list trend
Watch the rising-risk, high-CLV segment itself. Once the wiring fires, that list should shrink month over month as saves land before recency moves. If it holds steady or grows, rework the treatments, starting with the highest-CLV tier.
The predictive fields complement the scores you own. The RFM teardown gives you segmentation you can audit, with cut lines you set; the migration view shows whether any of this wiring moves people between buckets. The division of labor is the teardown's own framing: predictions as the check, RFM as the control.
Your account has been predicting this all along
When we read a qualifying account cold, the most common finding is this: every predictive field populated, recomputed weekly, and not one segment, flow, or campaign reading any of them. That read is part of the Blueprint. Or wire it yourself from this page; every step above runs on native conditions. Yours either way.
What operators ask
My account does not qualify. Now what?
Per Klaviyo's docs, the predictive fields need 500 or more purchasers and at least 180 days of order history. If you are under either line, the fields will arrive; sell and wait. Meanwhile, RFM scoring works from your very first export, no thresholds, and teaches the same instincts. Start with the RFM teardown.
How accurate are the predictions?
An established model family, retrained weekly per Klaviyo's docs, and still probabilistic. Some High-risk customers were coming back anyway; some Low-risk ones are gone. So wire the fields to cheap actions first, the automated touch and the refill nudge, and let results on your own account earn them bigger jobs.
Predictions or RFM: which should I use?
Both; they hold different jobs. The predictions are automated and opaque: the math runs weekly without you, and you cannot audit a score you disagree with. RFM is manual and transparent, with cut lines you own and can tune per brand. The teardown's FAQ gives the same answer from the other side.
The fields are live. Wire the first segment today.
Every wire on this page runs on native Klaviyo conditions; an afternoon covers it. Or book the fit call: the Blueprint includes this read, built in your Klaviyo and kept on exit. Yours either way.
More from this series: the lifecycle playbook the RFM teardown segment migration the Composer review
Sources
- Klaviyo: understanding predictive analytics
- Klaviyo: the churn prediction model
- Fader and Hardie: probability models for customer-base analysis
- BG/NBD model explainer
- Concomitant: the RFM scoring teardown
Platform specifics above are attributed to Klaviyo's documentation and will move as the product does; validate the behavior in your own account.