Date of Award

August 2020

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Engineering

First Advisor

Matthew E.H. Petering

Committee Members

Hamid Seifoddini, Christine T Cheng, Jaejin Jang, Wilkistar A Otieno

Keywords

Fully embedded optimization algorithm within simulation model, Online retail, Order delivery (last mile), Order fulfillment, Order fulfillment decision reevaluation, Supply chain management

Abstract

This dissertation makes both a methodological and an applied contribution. From a methodological standpoint, this is among the very first works in the literature to explore the concepts of true simulated operating conditions and fully embedded decision-making algorithms. We illustrate the effectiveness of these concepts by applying them to an online retailer (i.e. e-tailer) order fulfillment decision making process.

Online shopping has completely transformed retail markets in recent years. For customers, it provides convenience, visibility and choice, and for retailers it provides market expansion opportunities, operational cost reduction, and many other advantages. There are fundamental differences between the supply chain design and operations of an online and traditional (i.e. brick and mortar) retailer. One of the key differences exists in customer order fulfillment which refers to the process of picking and packing order items from a retailer’s warehouse or store and delivering them to customers. In traditional retail, order fulfillment happens in physical stores and by customers. In online retail, however, the tables are turned, and the retailer is responsible for this task.

The reliability, cost, and lead time of online order fulfillment have a direct impact on customer satisfaction and an e-tailer’s overall success. In today’s competitive market, excellence in fulfillment is critical and organizations are struggling with how best to accomplish this while remaining profitable. On one hand, order fulfillment accounts for a considerable amount of operational cost and reducing it directly improves an e-tailer’s bottom-line. On the other hand, customers demand fast and cheap order delivery options. This constantly pushes e-tailers to make tough strategic and operational choices to stay competitive.

An e-tailer’s order fulfillment process begins with a fulfillment decision which assigns a customer order to one or more fulfillment centers (FCs). E-tailers typically put an order fulfillment policy (i.e. fulfillment strategy) in place that determines how those decisions must be made. Identifying the best policy is extensively studied in the literature. However, most of the proposed policies focus on minimizing the fulfillment cost for individual customer orders by finding an optimal assignment at the time an order is placed. In this dissertation we show that this policy leads to a suboptimal decision at the system level. In other words, when a collection of these myopic fulfillment decisions is analyzed together, total fulfillment cost can be further reduced by optimizing the decisions for that group collectively.

Since e-tailers receive customer orders around the clock and at a fast pace, order fulfillment decisions are made automatically using an algorithm. Additionally, from an operational perspective, making fulfillment decisions on the fly for individual customer orders enables e-tailers to keep an updated available-to-promise inventory record for each stock keeping unit (SKU) and FC combination. It also allows them to provide an estimated delivery window to their customers in real time. Therefore, although in theory optimizing fulfillment decisions for a group of customer orders reduces costs, there are practical challenges in deploying this policy in a real-world e-tailer environment.

In order to address these challenges, we propose a reevaluation strategy that does not fully replace the automated order fulfillment decision making process. Instead, it periodically reevaluates and optimizes the fulfillment decisions for a group of orders that are waiting in the system to be processed and shipped to customers. We develop an integer programming-based reevaluation algorithm that can be triggered for a fixed number of customer orders or at regular time intervals. Our integer program considers several dimensions such as on-hand and on-order inventory, customer delivery preferences, shipping methods, and the number of boxes to minimize total fulfillment cost while maintaining the delivery time and service level for all customer orders. Additionally, since the large instances of the proposed model are mathematically difficult to solve to optimality, we develop a decomposition-based heuristic for those instances.

As noted, our proposed reevaluation algorithm must be triggered regularly during an e-tailer’s operations without interrupting other important processes relating to new customer orders, shipment of orders, and inventory replenishment. Therefore, in addition to reevaluation decisions, the computation time used by a reevaluation algorithm needs to be considered when designing an effective strategy. For example, for customer orders that need to be shipped on a given day, reevaluation decisions must be finalized before the shipping deadline.

To study the complex relationship between reevaluation and other processes, we embed our reevaluation algorithm inside a discrete event simulation model in such a way that both the decisions produced and computation time used by the algorithm are fed back to the simulation model. This novel method which was first presented by Petering (2015), enables us to study the tradeoff between the quality of the decisions produced and computation time used by the algorithm in order to recommend the overall best reevaluation strategy for an e-tailer according to its operational characteristics.

Finally, we conduct more than two hundred experiments in which the reevaluation algorithm is fully embedded in the DES model. The results confirm the effectiveness of reevaluation algorithm in reducing total fulfillment cost by an average of 5% for our test instances. It also illustrates the tradeoff between decision quality and computation time and allows us to perform scenario analysis to find the best overall reevaluation strategy for an e-tailer.

Available for download on Saturday, August 27, 2022

Share

COinS