Increasing cost-efficiency in online shopping deliveries
Increasing Cost-Efficiency in Online Shopping Deliveries
Professor Yehua Wei helps companies develop more flexible fulfillment networks.
When companies develop strategic plans for the future, they need to account for a high level of uncertainty.
That’s where Yehua Wei’s research comes in. “The general theme of my research is decision under uncertainty,” he says. “These decisions have a long-term impact, involve a lot of uncertainty, and many downstream tactical issues that you need to consider.”
Wei is an associate professor of business administration in the Decision Sciences area at Duke University’s Fuqua School of Business and his research has many e-commerce applications. He recently partnered with one of China’s leading online retail companies to investigate the possibility of adopting a more flexible fulfillment network.
The solution was described in the paper “Understanding the Value of Fulfillment Flexibility in an Online Retailing Environment” published in the journal Manufacturing & Service Operations Management. Wei collaborated with Levi DeValve, an assistant professor at The University of Chicago Booth School of Business, and researchers Di Wu and Rong Yuan, who were both scientists at Chinese e-retailer JD.COM at the time of the study.
Wei and his co-authors evaluated the company’s distribution system, which contains two tiers of distribution centers: the local distribution centers—smaller warehouses responsible for fulfilling demand from their local district—and the regional distribution centers—bigger warehouses that function like “parents” of the smaller ones, supporting them when they run out of products.
Currently, when a customer orders an item that is stocked out at the local distribution center, the bigger regional distribution center may be asked to fulfill that demand if they have that particular inventory.
The company was wondering if it would be a good idea to add more links to this distribution network to interconnect the smaller distribution centers. “What they wanted to find out was if they should consider allowing these smaller distribution centers to fulfill each other's orders,” Wei says. When the company described the problem, Wei saw an opportunity to use his expertise in network design to come up with a solution.
Adding this flexibility to the network also poses challenges, for example, increasing costs to maintain additional drivers, trucks, and routing software. So, Wei and his co-authors evaluated the option of a network with limited flexibility, where each local distribution center would be linked with only a subset of other local distribution centers. And they came up with a new policy to determine how the orders should be fulfilled within the reconfigured network.
Wei and his colleagues used a dataset provided by the online retailer of 378 products across six regions to simulate the operation of the new network under the new policy. The dataset included information like the location of the inventory, the fulfillment costs, and profits from each item.
The simulation revealed that the approach developed by the team could potentially improve the company’s profits on the order of tens of millions of U.S. dollars on an annual basis. “Ten percent of the improvement comes from implementing a smart policy compared with the default policy and 90 percent comes from the business adding the flexibility of the additional links between the local distribution centers,” Wei says.
Evaluating more complex networks
In this study, Wei and his colleagues evaluated a scenario with a limited number of distribution centers. “But what if there are 50 distribution centers? Then it might be a lot harder to determine where to add those links,” Wei says. In that case, you need to be more systematic and develop an algorithm to figure out how to add the links and solve this strategic network design problem.
To consider all the possible networks in such a scenario, you could easily get an astronomically large number, even bigger than the number of particles in the universe, Wei says. “If you consider that the computer takes one millisecond to evaluate one network, it would take more than the age of the universe.”
Along with colleagues Levi DeValve and Saša Pekeč, professor of decision sciences at Fuqua, Wei proposed a strategy to deal with these types of situations, which was described in another paper called “Approximate Submodularity in Network Design Problems.” “We figured out the greedy method, which basically adds the link that gives you the biggest gain each time, finding very effective networks,” he says. “It's a natural method that makes this problem computationally tractable.”
Wei notes that, while his work is often inspired by practical problems, the research often leads to interesting and relevant theoretical questions. “And once we find the answers to the theoretical questions, they sometimes also help us with other practical problems, leading companies to rethink their supply chain, network, and operations,” he says. “It is an ongoing cycle.”
This story may not be republished without permission from Duke University’s Fuqua School of Business. Please contact media-relations@fuqua.duke.edu for additional information.