DoorDash developed a two-stage causal machine learning framework to optimize promotional campaigns by estimating true incremental customer response and allocating offers under business constraints. Using Double Machine Learning (DML), they distinguish between customers who would order anyway versus those influenced by promotions. Case studies show the approach reduced cost per incremental order by roughly 50% for cross-category campaigns and improved order rate lift for restaurant discounts by personalizing both targeting and discount amounts. The framework treats promotions as discrete (who to target) or continuous (how much to offer) treatments, then optimizes allocation within budget limits to maximize incremental orders per dollar spent.
Table of contents
Why promotions are trickyHow we frame the problemOur frameworkCase Study 1: Targeting promotions for non‑restaurant deliveriesCase Study 2: Optimizing discount for restaurant consumersConclusion and future directionsStay Informed with Weekly UpdatesPlease enter a valid email address.Thank you for Subscribing!Sort: