Presentation on YouTube.

Impulse buying amounts to over 50% of the revenue in some retail stores. We analyze a large data set of customer receipts from a grocery store in Beirut, and develop a regression model that allows capturing "ripple effects," i.e., the change in traffic throughout the store resulting from any change in product allocation. This is then embedded within a mixed-integer nonlinear program that sets the shelf allocations across the store. We develop specialized linear approximations that yield high quality solutions and useful insights. For the store in Beirut, we anticipate a 65% improvement in impulse profit.