Crowdsourcing Last-Mile Delivery with Hybrid Fleets Under Uncertainties of Demand and Driver Supply : Optimizing Profitability and Service Level

Problem definition: With the emergence and growth of e-commerce, e-retailers are challenged to provide faster and more cost-effective last-mile delivery by crowdsourcing independent drivers (IDs). To mitigate the risk of insufficient ID supply, retailers such as Amazon, Walmart, and other platforms rely on a hybrid delivery fleet of having professional drivers (PDs) to complement IDs for more viable delivery services. Such hybrid fleet delivery systems involve complex planning and operational decisions under multiple sources of uncertainties, which imposes significant computational challenges. The potential value of using IDs may vary under impacts of the uncertainties. Methodology: We formulate the problem as a multistage stochastic integer program. At the planning level, we optimize the fleet size and allocation in each zone of PDs under uncertain demand and IDs' availability. On a daily basis, each stage corresponds to a delivery time window. For each time window, given a realized demand-ID scenario, we construct an expanded transportation network with time and delivery-capacity status to model the operations of the hybrid fleet including order allocation and routing. We develop an iterative method based on approximate dynamic programming that makes use of piecewise linear functions. Results: Via testing instances generated based on the City of Minneapolis, Minnesota, we demonstrate the efficacy of the solution approach and study the benefits of employing IDs and how the uncertainties impact the employment of IDs. Managerial implications: The hybrid delivery fleet can significantly increase the profitability and service level of last-mile delivery. The use of IDs can endure the impacts of larger demand surges and demand variation in spatial distributions. Given the lower-fixed-cost and higher-variable-cost structure for hiring IDs, increasing demand volatility can lead to less use of IDs and fewer system benefits in profitability and service level. More IDs are used (i) when the available IDs are more, (ii) when their vehicle capacity is larger, or (iii) when the demand locations are relatively far from depots/warehouses