Targeting retention offers with heterogeneous treatment effects

Causal ML
Uplift Modeling
Retention
A case-study template for deciding which customers should receive a retention intervention.
Published

April 26, 2026

Decision Question

Which customers should receive a retention offer if the goal is incremental retention rather than high predicted churn?

Why This Is Causal

The highest churn-risk customers are not necessarily the customers most likely to respond to an intervention. A targeting policy should prioritize treatment effect, not baseline risk.

Identification Strategy

Use experimental assignment when available. If the data are observational, state the conditional exchangeability assumptions and evaluate overlap carefully.

Candidate Methods

  • T-learner, S-learner, and X-learner baselines.
  • Doubly robust learners.
  • Causal forests for heterogeneous treatment effects.
  • Policy value estimation for comparing targeting rules.

Notebook Plan

Add notebooks such as:

  • notebooks/01-targeting-problem.ipynb
  • notebooks/02-uplift-baselines.ipynb
  • notebooks/03-causal-forest.ipynb
  • notebooks/04-policy-value.ipynb

Decision Translation

Replace this section with a concise description of who should be treated, who should not be treated, and how much value the policy is expected to create.