🔗 Uplift Models

Uplift Models#

Models that predict the incremental effect of treating each individual.

Important

✨ AI-generated content. This page was written with the assistance of an AI language model and is provided as a learning aid. Despite careful review, it may still contain mistakes, omissions, or out-of-date information. Whether you are new to the topic, a team lead, or a senior practitioner, treat it as a starting point rather than an authoritative reference: read it critically and independently verify anything you act on (code, commands, figures, and factual claims) against official documentation and primary sources before relying on it.

What it is#

Uplift models — also called incremental, net-lift or true-lift models — are predictive models that estimate the causal impact of an action (a campaign, discount or treatment) on an individual’s behaviour, rather than the behaviour itself. They shift the question from “who will buy?” to “who will buy because of the campaign?” — predicting incremental change, not raw outcome.

Four kinds of customer#

Uplift modelling divides people by how they respond to intervention: persuadables act only because of it (the target), sure things would act anyway, lost causes never act, and do-not-disturbs react negatively if contacted. The goal is to reach persuadables, stop wasting budget on sure things and lost causes, and avoid provoking do-not-disturbs.

Techniques#

Four families. The two-model approach trains separate treated- and control-group models and subtracts their probabilities. Class transformation relabels the target to encode both outcome and treatment, so a single classifier predicts uplift directly. Uplift trees and forests choose splits that maximise the treatment-control difference. And meta-learners (T-, S- and X-learners) build uplift on top of standard models within causal-ML frameworks.

Evaluation and trade-offs#

Because uplift concerns causal effect, accuracy and AUC are not enough; models are judged by the Qini curve and coefficient, the uplift curve, and AUUC. In a subscription campaign where the treatment group buys at 20% against a control’s 15%, the model tries to pinpoint which individuals make up that +5%. The payoff is better spend, ROI and causal insight; the costs are a need for experimental (A/B) data, greater model complexity, and harder interpretation.

Theme: Causal Inference & Uplift · All terminology

Hint

Mind map — connected ideas

Uplift · Uplift Score · Qini Curve · AUUC (Area Under the Uplift Curve) · Causal ML (Causal Machine Learning) · Treatment Effect

Hint

More in Causal Inference & Uplift

AUUC (Area Under the Uplift Curve) · Causal Effect · Causal Impact · Causal Inference · Causal ML (Causal Machine Learning) · Causal Trees · Cumulative Incremental Gain (CIG) · Cumulative Uplift · Incremental Conversions · Incremental Gain · Incremental Recovery Rate (IRR) · Incremental Revenue · Incremental Sales · Qini Coefficient