Three essays on on-demand service platform operations

Abstract

On-demand service platforms match demands from delay-sensitive customers with gig workers providing services. However, this new business creates new problems and challenges for platforms, gig workers, customers, and regulators. This thesis investigates the sustainable development of the platform economy from the perspective of on-demand service platform operations.

The first essay focuses on algorithmic regulation and worker insurance for on-demand service platforms. Platforms have been criticized for using algorithms to force gig workers to overwork and not providing them with insurance. Chinese regulators have issued guidelines that restrict the algorithmic allocation to reduce the work intensity and that call on platforms to offer worker insurance. Are workers indeed protected by algorithmic regulation and insurance? How do algorithmic regulation and insurance influence the customer surplus, the platform's profit, and social welfare? To answer these questions, we employ an equilibrium model that accounts for the interaction among price, wage, labor supply, work effort, customer delay, and demand. The main findings are as follows: (i) Algorithmic regulation increases labor welfare through the increment of the wage, which occurs when the population of workers is relatively larger and customers have high willingness-to-pay and low sensitivity-to-wait-time; (ii) Insurance, along with a wage floor, increases the labor welfare through the increment of customer arrival rate; (iii) Algorithmic regulation decreases the system's efficiency (i.e., increasing price and wage and decreasing the platform's profit, the customer surplus and the social welfare) while insurance increases the system's efficiency (i.e., decreasing price and wage and increasing the platform's profit, the customer surplus and the social welfare). This study helps regulators, platform managers, and gig workers understand how the efficiency of on-demand service systems can be improved or impaired.

The second essay examines the contract conflict between platforms and restaurants. Platforms usually charge a high commission together with many fees for restaurant owners. Motivated by the recent appeals (during the pandemic) from restaurants to remove or reduce commissions for online food ordering and delivery services, we study the contract design problem between a platform and a restaurant. Using a stylized game-theoretic model, we compare different contract schemes that involve only two out of the three contract terms: commission (which is proportional to the order value), service fee (a fixed fee for each order), and membership fee (a fixed fee for each restaurant). We find that charging the restaurant a membership fee plus a per-order service fee helps the platform achieve close to the maximum profit in most cases. The reason is twofold. First, the service fee transfers the fixed delivery cost to the restaurant and aligns the platform and restaurant’s interests. Second, the membership fee allows the platform to freely allocate profit between them. In contrast, a commission can neither align the platform and restaurant nor allow the platform to freely distribute profit. Hence, in most cases, charging a commission plus a service fee or a membership fee is sub-optimal for the platform. From a regulator’s perspective, it is socially efficient to ask the platform to waive the commission and cap the membership fee so that the platform and the restaurant can both obtain higher and positive profits.

The third essay investigates a worker-centric order assignment for food delivery platforms. To minimize the delivery time, platforms use algorithms to make forced order assignments and the shortest but unreasonable delivery routes for workers. Is it feasible to maintain delivery time while allowing workers to reject orders and plan their delivery routes? To answer this question, we propose a worker-centric assignment algorithm, which can be summarized as follows. First, we forecast the courier's real-time route plan by estimating a distribution of the courier's all possible route plans. The probability of a possible route plan is the product of the destination choice probabilities of each stage. We develop a tensor-based multinomial logit model to predict the destination choice. Second, we propose a greedy heuristic for assignment optimization. In this heuristic, the platform minimizes the sum of the expected average wait time and the expected delay rate for current outstanding orders. In the numerical simulation, we compare our greedy heuristic with several classic assignment strategies and find that our greedy heuristic has a significantly better performance.

These three studies provide new insights on on-demand platform operations from different aspects. By better understanding how platforms and workers make decisions during the service process, we can provide guidance to promote the sustainable development of the platform economy.