Method

Causal graph / DAG

Encoding assumptions in a DAG; backdoor criterion and adjustment sets.

11 cases from 9 companies apply Causal graph / DAG. 2 report results in production. Includes variants: Backdoor adjustment.

Typical tasks: Measure impact without an A/B test, Decide who gets an offer, Optimize decisions, Forecast demand and load and 4 more. Companies: Lyft, T-Bank, Sber, Skoltech, Walmart, DoorDash and 3 more. Often combined with: Backdoor adjustment, Double machine learning, Causal inference (general), Uplift modeling, Instrumental variables and 19 more.