Title: Models and Algorithms for E-Commerce Fulfillment Network Design

Committee:

Dr. Alan Erera (advisor), Industrial and Systems Engineering, Georgia Institute of Technology

Dr. Mathieu Dahan, Industrial and Systems Engineering, Georgia Institute of Technology

Dr. Santanu Dey, Industrial and Systems Engineering, Georgia Institute of Technology

Dr. Bistra Dilkina, Computer Science, University of Southern California

Dr. Benoit Montreuil, Industrial and Systems Engineering, Georgia Institute of Technology

Date and Time: Tuesday, April 23rd, 12-1:30 PM ET

Location: Groseclose 226A (Georgia Freight Bureau Conference Room)

Virtual Link: https://gatech.zoom.us/j/99165965516?pwd=RjFCbVRmT20wOXNlQVdiSks5SWZLdz09

Abstract:

Large e-retailers today must manage complex fulfillment networks to ship purchased products directly to customers. Products may be stocked in and shipped from retailer fulfillment centers (FCs) or they may be shipped directly from vendors. Depending on the shipment size, package transportation carriers (e.g., UPS or FedEx), or less-than-truckload (LTL) trucking firms may be used for shipping direct to customers. However, large firms may be able to generate substantial cost savings by alternatively building consolidated loads with many shipments outbound from some stocking locations into other facilities and potentially transferring those shipments into subsequent consolidated loads prior to last-mile delivery. Such a system of consolidated loads is a private middle-mile network, and the design of these networks is the focus of this thesis.

In Chapter 2, we study a middle-mile network design optimization problem with fixed origins and destinations to build load consolidation plans that minimize cost and satisfy customer shipment lead-time constraints. We propose models that extend traditional flat network service network design problems to capture waiting delays between load dispatches and ensure that shipment lead-time requirements are satisfied with a desired probability. To find high-quality solutions to the proposed MIPs, we develop an effective integer-programming-based local search (IPBLS) heuristic that iteratively improves a solution by optimizing over a smartly selected subset of commodities.

In Chapter 3, we propose an approach that leverages data on customer purchasing sensitivity to quoted order-to-delivery times (ODTs) when designing middle-mile consolidation networks to maximize the profit of e-commerce retailers. Our approach integrates quoted ODT-dependent sales volume predictions into a new MIP that simultaneously determines ODT quotes and a consolidation plan, characterized by the frequency of load dispatches on each transportation lane. The objective of the MIP is to maximize sales revenue net fulfillment cost while ensuring that quoted ODTs are met with a high probability as set by the retailer. Results from a U.S.-based e-commerce partner show that our approach leads to a profit increase of 10% when simply allowing a marginal change of one day to the current ODT quotes.

In Chapter 4, motivated by the aim to verify the quality of our heuristic solutions in Chapters 2 and 3, we develop bound-improving approaches for the fulfillment network design problem without time constraints. Most notably, we identify a new class of valid inequalities that produce stronger linear programming relaxation solutions as compared to previous work. We additionally perform an extensive computational study that demonstrate the value of our bound-improving strategies for fulfillment network design problems when using commercial optimization solvers.