Thesis Title: Demand and Capacity Management for Meal Delivery Systems 

Advisors: 

Dr. Alan Erera, School of Industrial and Systems Engineering, Georgia Tech 

Dr. Martin Savelsbergh, School of Industrial and Systems Engineering, Georgia Tech 

Committee Members: 

Dr. Alejandro Toriello, School of Industrial and Systems Engineering, Georgia Tech 

Dr. Pascal Van Hentenryck, School of Industrial and Systems Engineering, Georgia Tech 

Dr. Marlin Ulmer, Carl-Friedrich Gauss Department, Technische Universität Braunschweig

Date and Time:  May 25th, 2021 - 8:30 am EST

Meeting URL: https://bluejeans.com/538370228

Meeting ID: 538370228 

Abstract:

In meal delivery, customers place meal requests to restaurants, and a central platform coordinates the delivery process so that shortly after the prepared meal is delivered at a specified location. This service has experienced significant sales growth in recent years, both in the US and around the world. Operating a meal delivery network however is quite challenging, primarily due to the difficulty in managing the supply of delivery resources to satisfy dynamic and uncertain customer demand under very tight time constraints. In this thesis, we investigate demand and capacity problems that typically arise in meal delivery operations and solve them using optimization methods. We start with a general overview of meal delivery in Chapter 1, highlighting the development of the industry in recent years and its expectations in the near future, and reflect on the operational challenges that make meal delivery operations difficult.

In Chapter 2, we study several questions in meal delivery operations focused on matching the correct levels of supply with demand. To ensure excellent customer service, delivery aggregators may, for example, decide to temporarily decrease demand during an operating day by temporarily reducing the delivery area for one or more restaurants. We show that such simple demand restriction strategies allow a significantly smaller fleet to meet service requirements. To simplify the analysis, we focus on problem geometries that enable the use of stylized mixed-integer programs to optimally deploy a fleet of couriers serving large numbers of orders. Applying the proposed framework to several scenarios with one and two depots, we conduct an extensive experimental study of the effects on system performance of (i) allowing courier sharing between multiple depots, (ii) relaxing the delivery deadlines of placed orders, and (iii) restricting demand through limited adjustment of the coverage of restaurants. The results demonstrate the potential effectiveness of different dispatch control and demand management mechanisms, in terms of both the required courier fleet size to serve requests and the coverage level of orders.

Subsequently, in Chapter 3, we seek to fill a gap in the literature by considering, for the first time, the satisfaction of couriers in rapid delivery, which is critical in terms of retention/loyalty in a highly competitive environment. Under the premise that couriers prefer to operate in relatively small geographic areas to increase their efficiency, we propose the novel concept of dynamic courier regions: small operating regions for couriers which can also be dynamically and temporarily expanded to allow delivery capacity to be shared between regions when necessary to keep customer service performance metrics high. We propose an optimization-based rolling horizon algorithm for courier management that handles both region resizing and request assignment decisions. Experimental results for realistic settings demonstrate that the proposed algorithm successfully balances customer and courier satisfaction, simultaneously achieving delivery times that are comparable to those of a single operating region and courier satisfaction metrics that are comparable to those achieved by fixed, inflexible regions.

Lastly, in Chapter 4 we introduce and study the dynamic extra courier capacity acquisition problem, commonly observed in rapid delivery system operations. Delivery providers typically plan courier shifts for an operating period based on demand forecast. However, because of the high demand volatility, it may be necessary at times during the operating period to adjust and dynamically add couriers. We propose a deep Q-learning approach to obtain a policy that balances the cost of adding couriers and the cost of service quality degradation by an insufficient number of couriers. Specifically, we seek to ensure that a high fraction of orders are delivered on time and with a small number of courier hours. We present a computational study that shows that a learned policy outperforms policies representing current practice in the meal delivery space, demonstrating the potential of deep learning for solving operational problems in highly stochastic logistic settings.