Design an A/B test for promo-targeting models

Design an experiment to determine whether M1 outperforms M0 at deciding which users should receive a $5 promo. 1. Randomization and treatment delivery Randomize users 1:1 into two arms: control: ranked by M0. treatment: ranked by M1. Use a pre-registered, stratified random assignment based on pre-period activity, purchase frequency, and prior redemption history to improve balance. Each day, define the triggered eligible population identically for both arms: users who are…