Spring 2006 Seminar Series

DATE: Thursday, May 18, 2006
LOCATION: E40-298
TIME: 4:15pm
Reception immediately following in the Philip M. Morse Reading Room, E40-106

SPEAKER:
Diego Klabjan

TITLE
Inventory Models for Business-to-Consumer Fulfillment

ABSTRACT
Order fulfillment is vital for successful business-to-consumer e-commerce firms. For an e-tailer customers place orders from various geographically disperse locations, or, in the presence of multiple distribution channels, from various channels. We consider two different settings. Under the first model, a constant demand rate for each one of these markets in the continuous time infinite time horizon setting is assumed. The distribution cost for shipping goods from fulfillment centers to markets, the variable and fixed procurement costs, and the holding costs at the fulfillment centers are captured. Demand from a market can be fulfilled from several fulfillment centers. We first study the case of two centers and a single market. We show that only one of the centers is used by an optimal policy. If, in addition, each facility has a market that must use an assigned center, this might no longer be the case. An optimal policy might use one center for a period of time and the other one for a different period of time. We develop simple cyclic policies that might outperform the strategy of using a single center. We also study the general case of several centers and markets. We give properties of optimal policies and we develop a lower bound. To this end, we first develop a Lagrangian type lower bound for general weakly coupled semi-Markov decision processes and then we apply this bound to our problem. Computational experiments show that the lower bound is often tight. In the second model, we assume that each market has a stochastic demand in a single time period setting. We develop and analyze two approximate models. One of them can be analytically solved. In addition, we give a lower bound on the exact model that is used in the computational study. The solutions from the approximate models are reasonably close to the lower bound and they give an improvement over the strategy of assigning a unique fulfillment center to each demand location.