Full talk · edited transcript

Double Machine Learning vs Propensity Score Matching: How to Evaluate Non-Randomized Experiments Faster and More Reliably

Platon Popov23 May 2026

Platon Popov, Data Analyst on the A/B platform at Wildberries & Russ.

The author works on causal inference and variance reduction methods on the A/B platform. Talk outline: the case of advertising in pickup points (PVZ), why a t-test cannot be used there, how confounders were controlled via propensity score matching, why this method turned out to be unsuitable, the switch to double machine learning with better results, and the production pipeline.

Continue reading — it's free

Google or a one-time email code. No password.