The rise of online shopping has changed consumer expectations, with more people choosing to have products delivered directly to their homes. Grocery home delivery is a growing segment within this trend, but it presents specific logistical challenges. Groceries typically have low profit margins, while last-mile delivery is expensive. Orders are often large, heavy, and perishable, which means customers must be at home to receive them.
To reduce failed deliveries and increase convenience, online grocers allow customers to select a delivery time slot. This system helps coordinate deliveries, but it also creates a complex planning problem. Delivery companies must decide which time slots to make available, balancing customer choice with their own operational limits. If too many slots are offered, routes may be infeasible. If too few are offered, vehicles may leave capacity unused.
To manage this, e-grocers rely on acceptance mechanisms that dynamically close time slots for new customers when necessary. These mechanisms must ensure that all accepted orders can be delivered within the promised time windows, given the available fleet. However, determining feasibility is not straightforward. It essentially requires solving a Vehicle Routing Problem with Time Windows (VRPTW), a complex optimization challenge, for each new order and time slot. At the same time, customers expect near-instant confirmation, making speed essential.
A recent dissertation by Liana van der Hagen examines this issue by exploring how supervised machine learning can support time slot management. Instead of solving the whole routing problem for every request, predictive models can provide quick feasibility assessments, helping grocers decide whether to offer or close a time slot.
This research highlights how advanced analytics and AI can help online grocery retailers manage the delicate balance between customer convenience, cost efficiency, and operational feasibility in last-mile delivery.