Yanlu ZHOU - ESSEC Business School

September 02nd, 12pm - 1pm / room will be communicated soon - Campus Cergy

Market Thickness in Online Food Delivery Platforms: The Impact of Food Processing Times

Abstract: Online food delivery (OFD) platforms are expanding rapidly worldwide, allowing customers to order food from a wide array of restaurants on their mobile phones. The pace of this expansion is accelerated in part by changes in consumer preferences during the COVID-19 pandemic. We study new algorithms to match drivers with orders on OFD platforms - a core function of many of these platforms. We develop real-time matching algorithms that use the (highly variable) food processing time to `delay' the assignment of drivers to orders to thicken the `market' of orders and drivers. The policy relies on machine learning techniques to predict the food processing time of an order, and the dispatch timings are obtained through a careful trade-off  analysis of the supply-demand dynamics and the variability

in the food processing times. Traditional methods in this  field include greedy (online matching) and batching (matching in periodic intervals) policies. We introduce a new level-k policy to delay the assignment until we have no more than k matching options left, and we show numerically that this policy performs exceedingly well on real data. Numerical experiments based on a real dataset from Meituan show that our policy can reduce total costs by 42% compared to matching policies that are currently in use. To explain the advantage of the level-k policy, we evaluate its performance on a simplified model and derive several analytical properties - the total costs are quasiconvex in the level of market thickness, implying that an intermediate level of thickness is optimal. We also find that the food processing time is key information for the matching decisions. Finally, our  findings indicate that if the unit delay penalty cost is high, the decision-maker of the platform should match conservatively.