讲座: Efficient Real-time Policies for Revenue Management Problems in Online Retail
题目： Efficient Real-time Policies for Revenue Management Problems in Online Retail
讲座嘉宾：Yanzhe LEI, Ph.D. candidate in Technology and Operations at the Stephen M. Ross School of Business at the University of Michigan
There is little doubt that e-commerce retailing will continue to grow at a significant rate in the foreseeable future. Modern e-commerce retailers (e-tailers) deploy advanced analytics solutions to gain insights and maintain competitive advantages. In particular, three fundamental decisions that are usually made with the assistance of analytics are product display (which products to display in what order on search screens), pricing (dynamically adjusting over time), and fulfillment (where to dispatch individual items from). In this talk, we address the following question: how should an e-tailer make these three decisions jointly and dynamically in real-time such that its profit is maximized? Given the problem scale that e-tailers face in practice, solving for the exact optimal policy is computationally infeasible. Therefore, we propose a class of policies that is easy to implement and has provably near-optimal performance.
We first consider a special case of the dynamic optimization problem outlined above, where the e-tailer only needs to make pricing and fulfillment decisions. We propose a novel class of policies that works as follows: it first solves an approximated optimization problem before the selling season for baseline control parameters; during the selling season, it adaptively adjusts the baseline control parameters in real-time according to the realized demand. An important feature of the real-time adjustment is that it decouples the calculation of pricing and fulfillment decisions, making it computationally tractable. We show theoretically that the performance of this policy is close to that of the exact optimal policy. We further extend the proposed policy to the more general problem where the display decisions also need to be made. Specifically, we propose a tractable approximation formulation and novel randomization scheme that translates the solutions to the approximated problem into control parameters. Finally, we validate the performance of the proposed policy in large-scale numerical experiments with real data.
This talk is based on joint work with Profs. Stefanus Jasin, Amitabh Sinha, and Joline Uichanco from the University of Michigan, and Dr. Andrew Vakhutinsky from the Oracle Labs.
Yanzhe Lei is a Ph.D. candidate in Technology and Operations at the Stephen M. Ross School of Business at the University of Michigan, advised by Stefanus Jasin and Amitabh Sinha. Murray is broadly interested in the intersection between business analytics and operations management. In his research, Murray develops real-time prescriptive analytics solutions that are easily implementable in a dynamic business environment. Murray’s work lies in the area of pricing and revenue management, and is motivated by real business problems that arise in the context of e-commerce/omnichannel retail and service operations. Part of his research is supported by an NSF grant and is built on collaboration with Oracle Labs.