Glossary

Uplift modeling

Predictive modeling that estimates how much a treatment changes an individual's outcome, to target only the people it will actually influence.

Also called: uplift model, net lift modeling, persuasion modeling

Uplift modeling predicts, for each individual, the difference a treatment is expected to make to their outcome, rather than predicting the outcome itself or the probability they will respond. It splits any population into roughly four types: those who convert only if treated ("persuadables"), those who convert regardless ("sure things"), those who never convert either way ("lost causes"), and those who convert only if left alone ("sleeping dogs," whom a treatment can actively push away).

This differs from a standard propensity or response model, which ranks people by likelihood to convert and would happily target "sure things" who needed no treatment at all, wasting spend or risking annoyance for no incremental gain. Uplift models are trained on data from a randomized or quasi-randomized experiment, since estimating an individual-level causal effect requires comparing what happened under treatment against a valid counterfactual, typically approximated with matched treated and control examples.

Uplift modeling is used in marketing offer targeting, retention campaigns, and pricing, wherever treating everyone would waste budget on people it wouldn't move and risks provoking the "sleeping dogs" segment. It is a practical application of heterogeneous treatment effects estimation, and its accuracy depends heavily on having a genuinely randomized treatment assignment in the training data to begin with.

Last reviewed September 22, 2026

In the index now

Related terms

Heterogeneous treatment effects (HTE) Differences in a treatment's true effect across subgroups, showing an intervention can help some segments and harm or not affect others. Propensity model A predictive model that scores how likely an individual is to take a specific action, such as buying, churning, or clicking. Average treatment effect (ATE) The mean difference in outcome between treated and untreated units across a full population, the standard headline result of an experiment. A/B testing A controlled experiment that compares two variants to measure which one produces a better outcome. Customer segmentation Dividing a customer base into groups that share meaningful traits so each can be targeted or served differently.