PhD defence of Thomas Visser

Consumers and small businesses purchase more and more products online. Many of these products require the customer to be at home to receive the product. Examples of such products include furniture, appliances and groceries. Delivery failures, when the customer is not at home, are not only inconvenient for customer, but also very costly for the supplier: additional miles need to be made for new delivery and sometimes the product itself needs to be replaced. This also negatively impacts road congestion and emissions. Therefore, many online retailers offer their customers a number of delivery time slots at checkout. For instance, ‘Monday between 09:00 – 10:00’ or ‘Tuesday between 10:00 – 16:00’. When selected, the online retailer promises to deliver the purchased goods within the selected time interval. It turns out to be crucial to manage the availability of these time slots in real-time, but this can be challenging. Although methods have been previously investigated, they were never tested in a real-world setting. In his project, Visser worked with consultancy firm ORTEC (ortec.com) and the large online grocery retailer Albert Heijn Online (ah.nl).

In the first chapters of the dissertation, he focuses on a number of issues that arise when trying to implement such dynamic time slot management system in practice. He finds it is essential for such systems to have low computation times. This is because the computations are done in real time, and customers will be very unhappy to wait several minutes on the website to load. Visser investigates mathematical speed-up techniques and improve a relevant case in which delivery routes must be made with (historic) congestion.

Moreover, he sees issues arise when multiple customers are simultaneously interacting with the system, e.g. by placing an order or by requesting available time slots. Current state-of-the-art methods are not well suited to deal with this, leading to high customer waiting times on the website or high distances per order. Visser introduces new methods that are able to handle this. In our experiments, we simulate up to 8000 customers that interact with the system in real-time.

In the last chapter of the dissertation, he investigates a novel variant of dynamic time slot management, in which the customer can only choose from at most one time slot per day, dependent on the delivery location. This model further assumes routes are fixed 'apriori'. This is more in line with the business model of the online grocer Picnic (picnic.nl). Management of the time slot availability is much easier, but the challenge now is to determine good 'apriori' routes and time slot assignment. For this, we propose and compare different methodology.