Programação matemática aplicada à casos reais de designação de entregadores em uma empresa de food delivery

Abstract

With the COVID-19 pandemic, applications that provide food and product delivery services have become an essential activity for society. One of the challenges for companies that own these platforms, through which the customer orders meals and products, is the demand for online delivery services that have to be efficient and scalable. The objective of this dissertation is to develop a methodology for the problem of deliverymen assignment for orders made through online Food Delivery platforms, which can be categorized as an Order Assignment Problem. In the searched literature, the context under analysis is also referred as Food Delivery Problem. The proposed solution methodology encompasses an application that processes data obtained from real cases. In the sequence, a mathematical model implemented in Mixed Integer Linear Programming (MILP) can designate the best deliveryman for each order (in relation to time and distance criteria), taking into account the group of orders and deliveries that needed to be combined for that instant of time. The proposed model solves the cases in which the number of available couriers is equal or greater than the number of orders to be allocated, a condition that represents the majority of scenarios experienced by the company that provided the data for real cases. The model is also able to partially solve cases with the number of orders greater than the number of deliverymen. The results obtained by the proposed mathematical model are compared with real cases and suggest, on average, a lower delivery time (e.g., 30% to 50%) than the time actually spent, as a function of a more efficient allocation of the deliverymen. Scalability tests were also performed for the proposed mathematical model, which showed an increase in the model execution time as the number of orders and deliverymen increase. Based on the obtained results and on the system specialist validation, it has been concluded that the proposed methodology could be applied to real online scenarios and new possibilities for future related studies have been raised.