Full talk · edited transcript

The New Policy Hasn't Run Yet: Comparing It to the Old One Before an A/B Test

Ivan Yurashku23 May 2026

Ivan Yurashku, head of data analysis, Sber.

A talk on how to use off-policy evaluation (OPE) to pre-check ML decisions on historical logs, without running an A/B test. A model does not only predict an event or a probability, it also helps make decisions: which action to apply to which object. Such models often compete with each other, and we would like to compare them without an A/B test.

Continue reading — it's free

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