Uplift modeling estimates the incremental effect of a marketing action: P(outcome | treated) − P(outcome | not treated) at the individual or segment level.

Four archetypes matter: persuadables (respond to treatment), sure things (convert anyway), lost causes (never convert), and do-not-disturbs (treatment hurts). Budget should focus on persuadables.

T-learner, X-learner, and causal forests

T-learners fit separate outcome models for treated and control groups. X-learners improve efficiency when treatment is imbalanced. Causal forests discover nonlinear uplift patterns without hand-specified segments.