Optimal Experimentation for Learning Personalized Policies Across Locations
Participate
Information Systems and Operations Management
Speaker: Spyros Zoumpoulis (INSEAD)
Room Bernard Ramanantsoa
Abstract
Firms wish to learn personalized policies for customers in heterogeneous yet related locations to maximize their monetary gains. To do this, they conduct experiments at each location to estimate the parameters of a customer response function. A crucial decision is which action to assign to each participant in the experiment, especially when a participant can only be assigned one action or there are budget constraints. The existing experimentation methodology considers locations and experiments individually. In this work, we leverage the relationship between locations in the experimentation problem to learn more profitable policies by proposing novel estimators and a semidefinite programming approach.
Joint work with Georgina Hall and Stefanos Poulidis.