Design and evaluate an uplift model

You must target a 20% subset of customers with a free-delivery promotion to maximize incremental orders per dollar. Define the causal estimand, uplift(x) = P(Y=1 | T=1, X=x) − P(Y=1 | T=0, X=x). Propose: (a) an experiment to collect training data (randomized holdout, treatment density, compliance tracking), (b) a modeling approach (e.g., T-/S-/X-/DR-learner or direct uplift models) and key features while avoiding leakage, (c) offline evaluation metrics (Qini, AUUC, uplift-AUC) and cross-validation strategy, (d) on-policy decision rule for top-K targeting given a fixed budget and a per-user cost curve, (e) guardrails for fairness across zones and user tenure, and (f) how you’ll handle interference/spillovers (e.g., delivery capacity constraints, surge) and positivity violations. Finally, describe how you would deploy and run a sequential test to validate lift without bias from response saturation.