Generalized Synthetic Control for TestOps at ABI: Models, Algorithms, and Infrastructure.

The article focuses on the development and implementation of a novel optimization-based inference method called Generalized Synthetic Control (GSC) and its integration into TestOps, a large-scale experimentation platform at Anheuser-Busch InBev (ABI). GSC addresses key challenges in physical retail experiments—such as small treatment effects, noisy and nonstationary data, interference, and endogenous treatment assignment—that limit the effectiveness of traditional difference-in-differences (DID) methods. By constructing a synthetic control as an optimal convex combination of control units, GSC significantly increases statistical power (approximately 100×) and eliminates the need for restrictive assumptions like parallel trends. Deployed initially in Mexico and now expanding globally, TestOps enables ABI to draw statistically significant conclusions in about 65% of experiments compared to only 6% with DID, thereby identifying meaningful innovations that increase sales volume by 1%–2% and impact over $135 million in monthly revenue. The platform's design facilitates experiment configuration, tracking, and transparent inference for non-expert users, and the underlying methodology has broader applicability beyond retail, including healthcare and public policy domains.