What is predictive segmentation based on likelihood to purchase?

Predictive segmentation uses past behavior to estimate what a subscriber is likely to do next. Specifically, it scores each person on their probability of making a purchase, usually expressed as a 0-to-100 score or a percentile rank within your list.

The model looks at signals like email engagement, how recently and frequently someone has purchased, average order value, website behavior, and time on list. A machine learning model finds patterns across all of these that would be impossible to spot manually at scale. The output is a score you can segment against.

In practice, you'd create three or four tiers from those scores. High-likelihood subscribers (say, your top 20%) get targeted offers and time-sensitive incentives. Mid-range gets nurture content designed to move them closer to a decision. Low-likelihood either gets less frequent sends or enters a re-engagement flow. You're basically letting the model tell you where to spend your messaging budget.

You don't need to build the model yourself. Platforms like Klaviyo, Blueshift, and Salesforce Marketing Cloud have built-in predictive scoring you can turn on without touching code. They've already trained on enough data to produce usable predictions for mid-size senders.

The biggest caveat: these models need data to learn from. If you're sending a few hundred emails a month or have limited purchase history, the predictions won't be reliable yet. It's worth starting with engagement-based segmentation first, and layering in predictive scoring once your data volume is large enough to support it.