Sites Inria

Version française

Collaboration

4/05/2016

Colisweb refines its delivery model with Inria

© Colisweb

The problem Colisweb submitted to the Inocs team at Inria Lille - Nord Europe was a complex one.The aim was to work out the schedule for a fleet of delivery vehicles in advance - and without knowing their workload -so that the company could offer instant deliveries at no extra cost.

Foresee the unforeseen. That, in a nutshell, is the challenge that Colisweb set the Inocs team at Inria (common to ULB and Centrale Lille*). This Lille-based start-up has set out to provide guaranteed deliveries within Greater Lille within two hours of purchase, or by scheduled delivery in a one-hour delivery window, without substantially increasing its delivery fees. However, as every logistician knows, the "last mile" leg of a delivery is particularly costly and complex, since it involves juggling a fleet of delivery staff contending with the vagaries of city traffic and subject to the end-customer's availability... for a meagre profit margin! "On average, the time the delivery person spends collecting the parcel from its departure point and handing it over to the customer represents 40% of the time allocated to logistics " explains Thomas Pocreau, engineer at Colisweb. And this is precisely why retailers often have long delivery lead times and very wide delivery windows.

The complex problem of local delivery

Colisweb's promise sets the bar high. This budding Lille-based company rooted in the EuraTechnologies ecosystem offers two types of urban deliveries: either the delivery person reaches the destination within two hours of the online or in-store purchase, or a same-day appointment is made with the customer, who is guaranteed of receiving his or her parcel within a one-hour time slot. The founders' objective is also to give urban retailers the possibility of gaining a competitive advantage by offering their customers instant delivery. "Today, the costs and lead times involved are almost identical for a delivery from Paris to Lille as for a delivery from one point of Greater Lille to another. When you look at the routes deliveries have taken, you sometimes see that the parcel has travelled backwards and forwards a few times before finally reaching its destination. It's absurd! Retailers that are geographically close to their customers should be able to take advantage of this proximity to offer a very swift delivery service " explains Thomas Pocreau. To achieve this, Colisweb has formed a network of couriers, independent carriers and small transport companies available in the zone. Now it has to optimise the deliverers' routes and schedules in real time.

© Colisweb

Using mathematical programming to optimise workloads

The Colisweb system currently operates with a man-machine tandem. An algorithm makes an initial "naive" match, which is then validated by a dispatcher. Colisweb hopes to automate this phase with the help of software. The start-up has therefore approached the Inocs team at Inria Lille - Nord Europe and a partnership has been formed. Applied mathematicians Frédéric Semet, Inria's project lead, and Maria Isabel Restrepo are specialists in complex problem optimisation. "What makes this problem complex is the fact that, at the beginning of the day, no-one knows exactly what the delivery staff will have to do, and yet it still has to be planned out hour by hour " explains Frédéric Semet. Colisweb also wants to reduce delivery staff's stress levels by planning their day's work in advance. "When there's less pressure on couriers, they have fewer problems on the road. It's a proven fact. When they're calmer, they're also more likely to find solutions when they do run into a problem, such as a customer that doesn't answer the phone or an incorrect door code " points out Thomas Pocreau.

"To meet this challenge, we use constrained optimisation techniques in a stochastic framework " explain the researchers, "or in other words, optimisation in uncertainty. " The first module of their algorithm generates potential scenarios with probabilities of occurrence for the days and the week ahead. Colisweb can then assign timeframes and delivery zones to the delivery staff. For example, deliverer A is assigned the eastern zone of the capital from 8am to 12pm on Monday, and told that he or she will have to make four deliveries. "Then real-life conditions take over and change the scenario in real time, which is where the second module comes in " explains Maria Isabel Restrepo. The system then assigns specific deliveries to each delivery person. This workload optimisation allows Colisweb to reduce the delivery costs. " Inocs' algorithms also make it possible to take account of traffic conditions when managing a multimodal fleet of delivery staff and couriers (using bikes, scooters, cars, vans and even skateboards in Paris). The aim is to choose the best type of transport for each delivery. Colisweb hopes its customers will soon be able to track their deliveryman's progress in real time using a geolocation tool for smartphones.

Colisweb's success story

Founded three years ago, Colisweb has established its head office in Lille, in the 150,000m2 EuraTechnologies business park for new technologies. At its head are Rémi Lengaigne, a young engineer trained in entrepreneurship, and Damien Abgrall, already at the helm of a start-up specialised in IT development for the Web. The idea behind Colisweb stemmed from a "start-up weekend" in Lille: a sort of hackathon during which the candidates pitch their business plan to a jury. After this event, the two partners decided to bring their concept to life and set up a business. Now with a staff of 10, Colisweb runs operations in several large French cities. Its customers include such major retailers as Darty, Habitat and Leroy Merlin, among others.The start-up recently won the digital innovation prize awarded by Banque Publique d'Investissement.

* Within the joint research unit UMR 9189 - CNRS-Centrale Lille-Université Lille1, CRIStAL.

Keywords: Modelling Optimisation Transfer Start-up Mathematics Probability

Video gallery

Top