Method

Uplift modeling

Modeling incremental effect of an action per user (uplift trees/forests, two-model).

13 cases from 10 companies apply Uplift modeling. 5 were checked in an online A/B test; 3 report results in production.

Typical tasks: Decide who gets an offer, Measure impact without an A/B test, Optimize decisions, Predict churn and lifetime value and 2 more. Companies: VTB, Avito, Sber, Skoltech, Yandex Lavka, Duolingo and 4 more. Often combined with: S-learner, Uplift evaluation (Qini/AUUC), Backdoor adjustment, Meta-learners (S/T/X/R), Instrumental variables and 41 more.