Predictive analytics in Klaviyo: Using behavioral data to time and target email

Predictive analytics in Klaviyo generates four per-contact scores - predicted order date, customer lifetime value, churn risk, and time between orders - that retention marketers use to decide who receives which email and exactly when. The models update continuously as customer behavior changes, turning a static contact list into a dynamic map of intent and risk. For brands sending lifecycle email at scale, these scores are the difference between guessing at cadence and acting on signal.

A linguist’s framing of why predictive models matter

Anna Sophie Christensen, Head of Email Marketing & Retention at FABO and a Klaviyo Community Champion, traces her analytical instinct back to Spanish verb conjugation. “You have one verb and you can bend that in a thousand different ways based off of what sentence it is a part of,” she explains. The rules and dynamics of that complexity genuinely fascinated her, and she identifies that fascination as the starting point for the analytical mindset she now brings to email marketing.

That framing matters because Klaviyo’s predictive toolset is, at its core, a rules-and-dynamics problem. The platform’s machine learning estimates each contact’s next order date, forward-looking revenue contribution, and churn probability. Those predictions share the same structural shape Anna Sophie found compelling in Spanish: context-dependent, layered, and only useful if you understand what the model is actually doing.

Her background spans a bachelor’s and master’s degree shaped around both CSR and corporate communication, including technical work setting up a course platform. That combination of analytical precision, communication craft, and technical fluency is what makes her perspective on Klaviyo’s predictive features substantive rather than surface-level.

What Klaviyo’s predictive analytics actually models

Klaviyo’s predictive analytics generates four core scores for each contact, all of which update dynamically as customer behavior changes:

• Predicted order date: An estimate of when a customer is likely to place their next purchase, built on historical cadence and recent engagement signals.
• Customer lifetime value (CLV): A forward-looking estimate of the revenue a customer is expected to generate over the lifetime of the relationship.
• Churn risk score: A probability that a customer won’t purchase again, derived from behavioral signals including browse drop-off, email disengagement, and growing gaps between purchases.
• Average time between orders: The expected cadence between a customer’s purchases, used to set the right window for replenishment reminders, win-back flows, and inventory-prep emails.

What makes these scores useful for retention work is that they shift the question from “who is in this list?” to “what stage is this specific customer in, and what does that imply for what we send next?” Anna Sophie’s work at FABO sits precisely there: lifecycle stage as the organizing logic, with predictive signals as the inputs.

Why timing is the real unlock

Batch-and-blast email treats a 12-month-dormant customer identically to one who purchased yesterday. Predictive analytics breaks that pattern by making timing a function of individual behavior rather than a shared campaign calendar.

McKinsey’s 2021 research on personalization found that 76% of consumers are more likely to purchase from brands that personalize, and that companies executing personalization at scale can lift revenues by 5 to 15%. The math only holds, though, when timing is part of the personalization. A relevant email sent at the wrong moment in a customer’s cycle still underperforms.

Predicted order date enables what Anna Sophie’s work at FABO centers on: flows that activate based on where each customer sits in their buying cycle rather than where the brand sits in its content calendar. A customer predicted to purchase within 48 hours may need no promotional nudge at all. One whose churn risk score has crossed a meaningful threshold needs something different entirely - and probably something more substantive than a discount.

The distinction matters because it changes what “good email” means. Timing is a creative decision as much as a technical one. Anna Sophie’s analytical instinct - developed through years of working with systems where context determines form - is what allows her to use predictive outputs as inputs to actual flow design, rather than treating them as vanity metrics in a dashboard.

Churn risk as a retention signal

Churn risk scoring changes the retention brief meaningfully. Rather than running quarterly win-back campaigns to a static “lapsed customers” segment, teams using predictive analytics can act earlier - when the model detects behavioral drift before a customer has technically churned.

Salesforce’s State of the Connected Customer (2023) found that 88% of customers say the experience a company provides matters as much as its products. For retention teams, that reframes win-back: it’s a relevance problem more than a discount problem. The customer disengaged because something wasn’t working. Early churn signals give retention teams enough runway to actually diagnose that - rather than throwing 10% off at someone who stopped caring months ago.

Anna Sophie’s focus at FABO on trigger logic, timing, and lifecycle stages maps directly onto this: small decisions about when a flow activates, and what it says, compound over time. Predictive scores provide the data foundation for making those decisions deliberately rather than by default.

Behavioral data beyond the email list

One of the strongest inputs for Klaviyo’s predictive models is full-session behavioral data: what products a customer browsed, how frequently they return, what they add to cart without purchasing. Purchase history alone gives the model an incomplete picture. Richer behavioral signals sharpen the predictions.

This is where platform integrations matter in practice. Hello Retail passes behavioral signals - product views, search queries, recommendation interactions - directly into Klaviyo, so predictive models have session-level context beyond last-order date. For stores running both platforms, that means predicted order date and CLV scores are built on fuller data, which makes the lifecycle decisions those scores inform more reliable.

Epsilon’s research on personalization found that 80% of consumers are more likely to purchase when brands offer personalized experiences. That kind of personalization depends on having complete behavioral context, not just transactional history. The integration layer is what makes that context available to the predictive models that run the flows.

The instinct behind the tool

What makes Anna Sophie’s framing valuable is that she approaches Klaviyo’s predictive features as a practitioner who understands why the model works, not just how to switch it on. The Spanish verb analogy holds: the rules are interesting because they’re context-dependent. A predicted order date is only useful if the flow behind it accounts for what that timing actually signals about the customer. A churn risk score is only actionable if the retention flow it triggers has something genuine to offer.

Predictive analytics in Klaviyo creates the conditions for better email. The instinct to use those conditions well - to know what the model is detecting and why, and to design flows that respond to that signal appropriately - is still the marketer’s job. That’s the space Anna Sophie works in, and the reason her perspective on these features is worth following.

Watch the full Conversations episode with Anna Sophie Christensen: Email marketing beyond the blast.