Matching Supply with Demand

王譞 | ASADPOUR, Arash | ZHANG, Jiawei

Matching supply with demand is an important way to secure profits for businesses in many industries. However, supply–demand mismatch is common, because decisions about supplies are often made before demand is fully characterized. To address this problem, HKUST’s Xuan Wang and collaborators modeled how resource flexibility can be leveraged to reduce supply-demand mismatch.

“What we mean by flexibility,” say the researchers, “is the firm’s ability to make each resource ‘multifunctional’ and capable of serving multiple demand classes.” They offer the example of a customer service call center. If agents specialize in few inquiry topics, customers may experience long waiting times in the event of a demand spike; however, if agents have the expertise to handle multiple inquiry topics, waiting time can be effectively reduced. The latter is an example of how adding flexible resources to a system can help to overcome supply–demand mismatches.

Unfortunately, note the researchers, “it is usually prohibitively expensive or sometimes simply infeasible to implement a fully flexible system.” Dedicated systems—in which each resource serves only one specific demand class—are easier and less expensive to operate, but they tend to incur substantial losses due to misalignments between supply and demand.

Seeking a compromise, the team focused on partial flexibility via the “chaining” concept. In particular, they considered a long chain design,  a class of sparse networks in which each demand class can be served by two types of resources, which has been shown to perform almost as well as their fully flexible counterparts. “Fortunately,” say the researchers, “even just a small amount of flexibility, if configured in the right way, can be extremely effective in hedging against demand uncertainty and mitigating supply–demand mismatch.”

The effectiveness of this promising mitigation effect depends on the ways in which resources are allocated. The team developed a “modified greedy policy” that aims to balance the depletion of the inventory when selecting between two resource options. The team demonstrated that the long chain design, paired with the proposed modified greedy policy, can achieve similar performance as in a fully flexible system in terms of minimizing expected lost sales, regardless how large the market size is.

The team tested their system design and resource allocation policy with a real-world scenario—order fulfillment from an online retailer. Their strategy ranks the distance from a given resource and then sequentially assesses distances as pairs until a fulfillment request is completed. “For example, if both resources si and si+1 are out of stock when a type i request arrives, then we move down in the list and consider resources si-1 and si+2 and so on until the request gets fulfilled,” the authors explain.

This modified greedy policy outperformed previous benchmarking allocation strategies for outbound shipping costs for an online retailer. For two-day shipping, the savings ranged from 0.4 to 1.1%; for one-day shipping, the savings were 3–7%. These figures translate into important real-world benefits. “The outbound shipping costs for Amazon in fiscal year 2017 were about $21.7 billion,” say the researchers, “and a 1% reduction amounts to more than $200 million savings in shipping costs.”


Assistant Professor
Information Systems, Business Statistics & Operations Management