Klaviyo's Predictive Analytics has been quietly shipping for years. Most of the accounts we audit have it enabled by default and have never built a single segment or flow that uses it. That's a missed lever — these predictive signals are some of the cleanest first-party AI outputs available to an ecommerce email program, and they cost nothing extra to deploy if you're already paying for Klaviyo at a sufficient tier.
This post walks through what Klaviyo Predictive Analytics actually computes, how we use those signals as flow triggers and audience inputs across the Klaviyo email marketing practice, and the cases where the predictions are unreliable enough that you should ignore them.
What does Klaviyo Predictive Analytics actually predict?
For each customer profile with enough historical purchase data, Klaviyo computes:
- Predicted Customer Lifetime Value (CLV). Forecasted total spend over a future horizon.
- Expected Date of Next Order. The single most useful predictive field for flow timing.
- Expected Average Order Value. The forecasted order size at next purchase.
- Average Time Between Orders. Derived from purchase history; used to compute the next-order forecast.
- Churn Risk. A probabilistic score indicating likelihood of going dormant.
These are computed per profile, refreshed on Klaviyo's schedule, and exposed both in the profile view and as segmentation criteria.
Which predictions are reliable enough to act on?
Not all of them, equally. The rough hierarchy:
- Average Time Between Orders. Most reliable. It's a direct historical computation, not a model output. Use freely.
- Expected Date of Next Order. Reliable for customers with 3+ historical orders. Unreliable for customers with 1–2 orders.
- Predicted CLV. Useful as a relative ranking, less useful as an absolute dollar figure. Treat as cohort signal, not as a financial forecast.
- Churn Risk. Useful directionally; the threshold at which "high risk" becomes "actually churned" varies by category.
What this means in practice: build segments and flow triggers on the more reliable fields. Treat the less reliable fields as hints to layer on top.
How do I use Expected Date of Next Order as a flow trigger?
This is the highest-leverage use of Klaviyo's predictive stack — and the one most accounts miss.
The pattern:
- Build a flow triggered by "Expected Date of Next Order is within X days." The X depends on your category — 7 days works for consumables, 14–30 days for considered-purchase categories.
- Segment the flow on profile properties. First-time buyer vs repeat buyer get different messaging. VIP cohort gets different messaging than the broader engaged list.
- Sequence the flow with three emails over the predicted window. Reminder, social proof / new arrivals, final incentive.
This works because it sends the right message at the moment the recipient is statistically most likely to repurchase — not on an arbitrary calendar cadence.
What about Churn Risk as a winback trigger?
Useful, with a caveat. Klaviyo's churn risk signal often fires earlier than you'd intuitively define "lapsed" — which is good for proactive winback but creates a sequencing question against your existing Winback flow.
The pattern we use:
- High Churn Risk → pre-churn save flow. Different message than the standard winback. "We noticed you haven't been around — here's what's new." Lighter incentive, more product-discovery framing.
- Lapsed (no purchase in 90+ days) → standard winback flow. As described in our Klaviyo flows post.
- Don't run both flows on the same profile. Set the standard winback to skip profiles already in the high-risk save flow.
Sequenced this way, the predictive signal catches at-risk customers earlier and gives the standard winback flow cleaner cohorts to work with downstream.
How do I use Predicted CLV for segmentation?
Treat predicted CLV as a cohort ranking, not a forecast.
The minimum:
- Top 10% by predicted CLV → VIP cohort. Different campaign cadence, different offer logic, often earlier-access creative.
- Top 11–30% → high-value cohort. Standard cadence with margin-friendly offers.
- The rest → engagement-based segmentation. Treat predicted CLV as a secondary signal.
The CLV cohorts are also useful as value-based bidding signals in paid media. Export the top-cohort segment to Customer Match in Google Ads and Matched Audiences in LinkedIn; use as seed audiences for lookalikes in Facebook advertising. The same first-party data that powers email segmentation also improves paid acquisition quality.
When should I ignore the predictions?
Three cases:
- Low data volume profiles. Customers with 1–2 historical orders have predictions that are essentially defaults. Don't trigger flows off them. Wait for the third order, then they become predictive.
- Catalog launches with no historical analog. A brand launching a new product category has no purchase history for that category. Klaviyo's predictions extrapolate from prior categories; they're often wrong for genuinely new launches.
- Seasonal anomalies. Holiday-driven purchases skew the next-order date forecast. The model doesn't fully decompose seasonality. Layer your own seasonal logic on top.
In all three cases, fall back to engagement-based segmentation and the deterministic flow triggers (abandoned checkout, post-purchase, etc.).
How does this fit alongside Klaviyo AI for content?
Klaviyo's AI also produces subject line suggestions, send-time recommendations, and (recently) generative copy assistance. Those are content-layer tools; Predictive Analytics is a data-layer tool. They're separable:
- Predictive Analytics → who and when.
- Klaviyo AI for content → what and how.
We use Predictive Analytics aggressively. We use the generative content features sparingly — generated subject lines and copy benefit from human review before they ship to the audience, and brand voice is a thing that AI generators degrade if you let them run unsupervised.
How does this stack with paid media targeting?
The Klaviyo predictive segments are useful exports for paid media:
- High predicted CLV cohort → seed audience for Facebook advertising lookalikes.
- At-risk / pre-churn cohort → retargeting audience for PPC services display campaigns.
- Expected-next-order-within-7-days cohort → exclusion audience for prospecting (don't pay to acquire someone about to buy organically) AND inclusion for retargeting (a well-timed retargeting touch may accelerate the purchase).
The audience export isn't automatic — you turn it on per destination in Klaviyo's integration settings. Worth the setup time.
Key takeaways
- Klaviyo Predictive Analytics computes predicted CLV, expected next order date, AOV, time between orders, and churn risk per profile.
- Expected Date of Next Order is the highest-leverage signal. Build a flow triggered on it.
- Churn Risk catches at-risk customers earlier than a calendar-based lapsed definition. Use it as a pre-churn save flow, sequenced before standard Winback.
- Treat Predicted CLV as a cohort ranking, not a forecast. Use it for VIP segmentation and as a paid-media seed audience.
- Ignore predictions for low-data-volume profiles, new-category launches, and seasonally anomalous periods.
- The predictive segments are useful exports for Customer Match, Matched Audiences, and lookalike seed audiences across paid media.
Want a Klaviyo audit?
If your Klaviyo program is running, but predictive segments and triggers aren't pulling weight, that's leverage on the table. The 1Digital® ecommerce email marketing team audits Klaviyo programs against this framework — tell us about your account to start.