What is Propensity Modeling?

Propensity Modeling assigns each customer a probability that they will perform a specific action, such as buying a category, converting on a campaign, churning or responding to a discount. In a composable storefront these scores live in the customer data platform and are read by the decision engine, the recommendation engine and edge personalization workers to pick the right experience in real time.

Definition

Propensity Modeling uses supervised machine learning trained on first-party data and identity-graph-stitched behavior to estimate the likelihood of a future event. Features come from observed activity, transactional history, Zero-Party Data and cohort membership; labels come from historical outcomes. Scores are recomputed on a schedule and on demand, then published as profile attributes that the storefront can query at session start. Consent-bound usage rules ensure that scores are only applied to channels and contexts the customer has agreed to, even when the model itself trained on aggregated data.

Why it matters

Personalization without scoring is a series of guesses that depend on hand-written rules. Propensity scores compress thousands of weak signals into a small set of strong, comparable features that a decision engine can rank. They power next best offer selection, audience segmentation refinement and triggered marketing automation. In a composable setup propensity outputs sit at the boundary between data and experience: the CDP computes, the experience layer consumes, the commerce service stays focused on transactions. This separation lets teams retrain models without redeploying the storefront.

Use cases

A fashion retailer ranks PLP items by a category-specific buy propensity score so the most likely purchases sit above the fold. A subscription brand pipes a churn-prediction propensity into checkout, triggering a retention offer when the score crosses a threshold. A B2C marketplace uses upsell propensity to choose between two cross-sell modules in the cart, switching between bundle and single-item dynamic content. An email-marketing system uses send-time propensity to schedule deliveries, lifting click-through rate without changing the creative.

Related

Explore Agentic Frontend Management Platform · Personalization.