Machine learning for time slot management in grocery delivery

Online shopping has grown in popularity in recent years, giving consumers the opportunity to have products delivered to their doorsteps. Grocery home delivery, in particular, presents logistical challenges due to the low profit margins of grocery products and the high last-mile delivery costs. Since grocery orders are often bulky and perishable, customers need to be at home to receive their groceries. To prevent delivery failures when customers are not at home to receive their orders, and for customer convenience, online grocers typically let customers select a delivery time slot to receive their orders.
To make effective use of their delivery capacity, online grocers can dynamically close time slots for certain new customers. E-grocers need an acceptance mechanism to manage the availability of time slots given their vehicle fleet. They want to ensure that there exists a feasible route schedule for which all accepted orders can be delivered in the selected time slots. However, offering too few time slots can be inefficient, leaving delivery capacity in the vehicles unused. Additionally, customers expect to receive the time slot offer almost immediately. Performing fast and accurate feasibility checks is challenging as it involves solving a Vehicle Routing Problem with Time Windows for each customer and time slot.
In this dissertation, we study time slot management in online grocery retailing, focusing on the complex relationship between order acceptance and delivery route planning. We explore the use of supervised machine learning to support time slot decisions in this context by providing fast feasibility predictions.