TITLE: Data-Driven Retail Revenue Management

SPEAKER: Kris Johnson

ABSTRACT:

This talk highlights two projects in a stream of work that uses data-driven approaches to develop effective revenue management strategies for online retailers.

In the first project, we partner with the online flash sales retailer Rue La La to develop and implement a pricing decision support tool that sets initial prices for new products. To do this, we use machine learning techniques to predict future demand and develop an algorithm to efficiently solve the subsequent multi-product price optimization. We show results from a five-month long field experiment that we designed and conducted to test the impact of our tool’s recommended price increases on demand, and we estimate an associated impact on revenue of approximately 10%. This is joint work with David Simchi-Levi and Alex Lee and was awarded the 2014 INFORMS Revenue Management and Pricing Section Practice Award.

Motivated by our work with Rue La La, our second project considers the dynamic pricing problem of an online retailer facing limited inventory and unknown demand. The multi-armed bandit problem has been a well-studied formulation of the dynamic pricing problem given unlimited inventory of a product. We have incorporated limited inventory constraints into a common algorithm for the multi-armed bandit problem, and we show that our algorithm has strong empirical and theoretical performance results. This is joint work with David Simchi-Levi and He Wang.