FoxyFoodDelivery logo FoxyFoodDelivery: When food complexity logistics become clear.

Context

In the US the restaurant industry market is worth $500 billion, whereas food delivery represents $30 billion. According to financial analysts (Morgan Stanley Inc.) the food delivery industry is a growing market. It could potentially reach $200 billion in few years. This behaviour is generally observed in developed countries: people want to eat food prepared by restaurants at home.

There exists a lot of competitors sharing this huge market, from start-up to well established company, but they all share a common point: the whole process, from clients ordering on their platforms to clients receiving their orders, is not autonomously treated. The part of human interactions, at many different levels, is important and leads to decisions that are not optimal. Consequently this poor organisation leads to a waste of time, and so the cost of the delivery is still important, for instance discouraging single people to order.

Our goal is to completely remove the human decision factor. Because there exist so many constraints that a human is not able to deal with optimally. To solve this problem we are creating a numerical platform that will be our decision maker. Our first modelling is to cluster restaurants, and then numerous algorithms inspired by Vehicle Routing Problems and Multi-Objective Optimization organise the full process of delivery.

Overview

Customer ordering meals using their dedicated application can see the global situation around them, thanks to real time information provided by the platform (see Fig.1). We present them possible already clustered restaurants, so they can join the community of buyers for each cluster. This gives new opportunities for customers, they are not restrained to order into one restaurant only, then can compose their meals with dishes coming from different restaurants. The cost of delivery is of course distributed around customers, because there is only one deliveryman associated to the cluster of restaurants.

Global situation when a new client opens his application.
Figure 1: Global situation when a new client opens his application.

Contribution of TaM

In order to produce a turnkey solution, as there are many physical actors (client, restaurant, deliveryman) each actor must have a dedicated application entirely developed by TaM team. (see Fig.2).

FoxyFoodDelivery App.
Figure 2: a) Client's App. b) Restaurant's App. c) Deliveryman's App.

The link between them is made by the creation of a numerical platform (see Fig.3). This platform contains not only the modelling of the full food delivery chain but also a set of specific algorithms (from positioning to optimization). The combination of them creates an optimal decision maker that is autonomous and takes all decisions according to a huge number of parameters.

Platform communication graph flow.
Figure 3: Platform communication graph flow.

Innovation

Modelling the full food delivery chain: It consists in expressing in a formal language very diverse data (physical, economical, human, ecological) that obey many constraints. We have to select the relevant data then to model their associated constraints, in order to create a coherent and realistic model. There are of course obvious data to model, like the paths that the deliveryman has to follow, but how to model a dish that contains variables like its preparation time, how fast it gets cold, its size, etc.? And how to take that into account when computing the instructions given to the deliveryman or to the restaurant? We are modelling the full food delivery chain from a general point of view, so our model could apply to any kind of existing food delivery company.

Example of timeline dealing with dishes conservation times during the delivery.
Figure 4: Example of timeline dealing with dishes conservation times during the delivery.

Specific algorithm creation: Routing problems form a highly-studied family of problems since 1959, they typically involve combinatorial and optimization techniques. These problems are frequently used to model real cases, but they are formulated in terms of single objective minimization, always minimizing the cost of the solution, depending on some constraints. However real life problems encountered in industry are fundamentally multi-objective. These objectives arise from different constraints: physical, economical, human, ecological. The use of multi-objective optimization to create models is recent from a research point of view, it has begun in the 90s, and is still in development. As far as we know there doesn’t exist a formulation adapted to the needs of food delivery industry. We are designing a specific formulation of the problem according to our modelling of the full food delivery chain.

Algorithm  process to generate a cluster of restaurants according to real time constraints.
Figure 5: Algorithm process to generate a cluster of restaurants according to real time constraints.