The Limits of Centralized Pricing in Online Marketplaces and the Value of User Control
May 10, 2022Springer book chapter on “Nonparametric Estimation of Choice Models”
September 26, 2022
Our work done in collaboration with AAnheuser-Busch InBev (AB InBev) and Arena AI was selected as a finalist for the 2022 Revenue Management and Pricing Practice Award. We implemented a bandit-based active learning approach to optimize manufacturer-to-retailer promotions at AB InBev, which is a $50B+ beverage multinational.
In a real-world A/B test conducted on a single SKU and a sample of 19K stores in the Brazil market, the active learning approach outperformed existing promotions strategies obtaining 39% more profit. Our innovation lies in the use of active learning to overcome
historical data challenges and a probabilistic and batched UCB algorithm to deal with complex business environments. A working paper on this work is forthcoming.
