Sheng Liu

Will Hindsight Approach Work Well for Multi-Location Inventory Allocation? Online retailers often decide where to place inventory before knowing which customers will arrive and in what order. A common solution is the hindsight or stochastic programming approach: allocate inventory as if future demand were known, then fulfill orders using a simple greedy rule, shipping from the cheapest available warehouse. In “Inventory Allocation under the Greedy Fulfillment Policy: The (Potential) Perils of the Hindsight Approach,” Jasin, Liu, and Zhao show that this widely used method can be deceptively optimistic. The hindsight solution is equivalent to optimizing greedy fulfillment for the most favorable arrival sequence, rather than the sequence that unfolds. The authors characterize when this optimism is harmless—identical ordering costs, no warehouse capacity constraints, and independent demands—and show that violating any of these conditions can make the hindsight allocation arbitrarily worse than the true optimum under greedy fulfillment. Their results caution retailers to evaluate allocation heuristics against the fulfillment policy actually used.

Rethinking How Cities Prepare for Extreme Rainfall Cities often prepare for extreme rainfall using a single “design storm,” such as a 100-year storm of a predetermined duration. But storms with the same return period can produce very different flooding losses. In “Planning Stormwater Retention for Resilience Against Extreme Rainfalls,” Liu, Qi, and Zhang develop a robust optimization framework that identifies the most damaging rainfall scenario for a city based on its existing drainage and retention infrastructure. Case studies of Toronto, Manhattan, and New Orleans show that the storm a city should prepare for changes with local infrastructure and climate conditions. The analysis also challenges the idea that greener is always more resilient; although green infrastructure offers important environmental benefits, concentrated gray infrastructure can become more valuable when rainfall is severe or budgets are tight. In Toronto, robust planning could save an estimated $6 million annually today and as much as $1.5 billion annually by 2050.

On-demand delivery has emerged as a fast-growing market, as customers crave speed for their purchases. Through an analysis of Meituan’s data, we find that a service region is often divided into delivery zones, and couriers may or may not have the flexibility to serve demand across zones. Motivated by this observation, this paper investigates whether providing dispatching flexibility to couriers is beneficial and, if so, how to improve delivery efficiency by tailoring the flexibility design to the market. We model the on-demand delivery system with and without batching using queueing approximations. The expected customer wait time is used to measure the performance of different systems. We find that the fully flexible delivery system yields the best performance for single-order delivery. However, full flexibility may not be beneficial for batch delivery. We prove that the dedicated delivery system outperforms the fully flexible delivery system when every courier delivers two orders in high-traffic regimes. Furthermore, we propose a partially flexible delivery system in which only a fraction of couriers are flexible enough to serve multiple zones, and we show that it can perform better than both dedicated and fully flexible delivery systems. We perform simulations with synthetic data and a case study using real-world data from Meituan, both of which corroborate our theoretical findings. Our study reveals the advantages and limitations of dispatching flexibility in on-demand delivery. Pooling couriers to serve all zones enhances system responsiveness and adaptability to order arrival uncertainty, but carries the risk of inducing increased delivery distances and slower service processes with batch delivery. Interestingly, the proposed partially flexible delivery system demonstrates significant potential to improve operational efficiencies while reducing the potential hassle of operating a fully flexible system. History: This paper has been accepted for the Transportation Science Special Issue, “The First INFORMS TSL Data-Driven Research Challenge.” Funding: H. Wang acknowledges the support from the Start-up Fund for RAPs under the Strategic Hiring Scheme at PolyU [Grant P0058278]. S. Liu acknowledges the support from the Natural Sciences and Engineering Research Council of Canada [Grant RGPIN-2022-04950]. S. Wang acknowledges the support from the National Natural Science Foundation of China [Grant 72371221], and the Research Grants Council of the Hong Kong Special Administrative Region, China [Project HKSAR RGC TRS T32-707/22-N]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/trsc.2025.0095 .

How should retailers and sellers distribute inventory across a large network of (potentially hundreds of) distribution centers? Motivated by the emerging challenges of inventory allocation facing volatile demand, we develop distributionally robust network inventory management policies with cross-fulfillment using a scenario-based approach that is adaptable to feature information and demand correlation. We analytically demonstrate the value of scenario modeling for a two-location newsvendor problem. The key technical development is a computationally efficient linear programming formulation for large networks, which we further incorporate into a dynamic inventory policy combined with look-ahead approximation. We validate the superior performance of the proposed policies using real-world transaction data from a leading logistics service provider and a major retailer in China. The developed policy outperforms existing data-driven approaches and can save up to 19% of operating costs over the current policy employed by the company.

Problem definition: Parcel logistics companies use zoning systems to manage last-mile delivery operations. This practice divides a service area into zones, each served by its own station and drivers. Designing an optimal zoning policy is challenging because the practical service area often includes many customer locations, and vehicle routing problems (VRPs) should be incorporated as a subroutine. Existing methods have limitations in modeling practical fleets with diverse vehicle types and broader routing objectives. Methodology/results: We collaborated with a delivery company to develop a novel data-driven zoning method that minimizes the maximum work span of delivery stations. We define the work span of a station as the duration between the start time of sorting the first parcel and the return time of the last driver upon finishing all assigned delivery tasks. Our method iteratively solves VRPs using observed demand data and partitions the region with additively weighted Voronoi diagrams. We leverage the primal-dual properties and develop a subgradient algorithm with established convergence conditions. Our numerical analyses show that this approach not only reduces the station-level maximum average work span by 20.5% and the average delivery time per driver by 17%, but it also reduces their standard deviations by approximately 25% and 19%, respectively. When tested in actual field conditions, we continue to observe reductions in the work span of the stations and the delivery time of the drivers. Managerial implications: Our approach reduces delivery lead times, better distributes workload among drivers, and limits long working hours, creating a win–win outcome for both the company and its drivers. Besides improving service quality and driver well-being, we estimate annual savings of nearly half a million dollars simply by readjusting the boundaries of service zones. The proposed framework can also be applied in other spatial service settings to achieve equitable distribution of workload among resources. Funding: J. G. Carlsson acknowledges funding support from DARPA [Grant HR0011-25-3-0234] and USDOT [Grant USC-DOT-1217]. S. F. W. T. Lim acknowledges funding support from the Eli Broad College of Business 2024 Summer Research Grant. S. Liu acknowledges funding support from the Natural Sciences and Engineering Research Council of Canada [Grant RGPIN-2022-04950]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2024.1221 .

1 IntroductionThe booming popularity of on-demand delivery services prompts the on-demand supermarkets and grocery retailers, including Sam’s Club, China Resources Vanguard, Freshippo, and Whole Fo...

Problem definition: How should retailers leverage aggregate (category) sales information for individual product demand prediction? Motivated by inventory risk pooling, we develop a new prediction framework that integrates category-product sales information to exploit the benefit of pooling. Methodology/results: We propose to combine data from different aggregation levels in a transfer learning framework. Our approach treats the top-level sales information as a regularization for fitting the bottom-level prediction model. We characterize the error performance of our model in linear cases and demonstrate the benefit of pooling. Moreover, our approach exploits a natural connection to regularized gradient boosting trees that enable a scalable implementation for large-scale applications. Based on an internal study with JD.com on more than 6,000 weekly observations between 2020 and 2021, we evaluate the out-of-sample forecasting performance of our approach against state-of-the-art benchmarks. The result shows that our approach delivers superior forecasting performance consistently with more than 9% improvement over the benchmark method of JD.com . We further validate its generalizability on a Walmart retail data set and through alternative pooling and prediction methods. Managerial implications: Using aggregate sales information directly may not help with product demand prediction. Our result highlights the value of transfer learning to demand prediction in retail with both theoretical and empirical support. Based on a conservative estimate of JD.com , the improved forecasts can reduce the operating cost by 0.01–0.29 renminbi (RMB) per sold unit on the retail platform, which implies significant cost savings for the low-margin e-retail business. History: This paper has been accepted as part of the 2023 Manufacturing & Service Operations Management Practice-Based Research Competition. Funding: This work was supported by the National Natural Science Foundation of China [Grant 71991462]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2022.0453 .

We develop and implement a novel zoning optimization framework for Ninja Van, a major last-mile logistics provider in Southeast Asia. The framework integrates weighted Voronoi partitioning with vehicle routing under uncertain demand and varying vehicle capacities. Deployment of the system led to measurable gains in efficiency and equity, reducing delivery times and improving workload balance across stations.

Urban infrastructure is vital for sustainable cities. In recent years, municipal governments have invested heavily in the expansion of bike lane networks to meet growing demand, promote ridership, and reduce emissions. However, reallocating road capacity to cycling is often contentious because of the risk of amplifying traffic congestion. In this paper, we develop a method for planning bike lanes that accounts for ridership and congestion effects. We first present a procedure for estimating parameters of a traffic equilibrium model, which combines an inverse optimization method for predicting driving times with an instrumental variables method for estimating a commuter mode choice model. We then formulate a prescriptive model that selects paths in a road network for bike lane installation while endogenizing cycling demand and driving travel times. We conduct an empirical study on the City of Chicago that brings together several data sets that describe the urban environment—including the road and bike lane networks, vehicle flows, commuter mode choices, bike share trips, driving and cycling routes, demographic features, and points of interest—with the goal of estimating the impact of expanding Chicago’s bike lane network. We estimate that adding 25 miles of bike lanes as prescribed by our model can lift cycling ridership from 3.6% to 6.1%, with at most a 9.4% increase in driving times. We also find that three intuitive heuristics for bike lane planning can lead to lower ridership and worse congestion outcomes, highlighting the value of a holistic and data-driven approach to urban infrastructure planning. This paper was accepted by Karan Girotra, operations management. Funding: Funding: The authors acknowledge funding from the UCLA Anderson Easton Technology Management Center (Siddiq & Zhang) and the Natural Sciences and Engineering Research Council of Canada [Grant RGPIN-2022-04950] and TD-MDAL Research Grant from the Rotman School of Management (Liu). Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2022.00775 .