How to best estimate the number of students who will have lunch in the canteen on a given day when food orders must be made several weeks in advance? The answer to this question is all the more important since it concerns around 15,000 meals served every day in the 115 schools in Nantes.
find innovative solutions
The city of Nantes has therefore launched a call for tenders to find innovative solutions capable of responding to this problem. “The objective was to show ‘how can we use AI to serve the efficiency of a public policy? We could very well have chosen the mobility sector“, explains Franckie Trichet, deputy mayor of the city of Nantes and vice-president of the Metropolis, to The Digital Factory.
The start-up Verteego, specializing in predictive analysis, was selected alongside Maestria, a specialist in the deployment of AI. Both are members of NaonedIA, a Nantes collective that promotes ethical AI and serving citizens. The objective: to develop a predictive algorithm to reduce the rate of food “overproduction”.
A 9.6% rate of food overproduction
In practice, there is about 9.6% of food “overproduction” in Nantes canteens. A rate to be distinguished from that of wastage which corresponds to waste. This overproduction takes the form of products not consumed but which are not thrown away. In Nantes, they are redistributed to associations but still represent a significant financial cost, “almost impossible to fix“, indicates Franckie Trichet.”A single meal costs 1.65 euros on average. But this amount is in fact higher because it does not take into account all the upstream logistical effects and the environmental consequences “, adds the deputy mayor.
Thus, in collaboration with the city’s business teams, the central canteen and Maestria, the start-up Verteego has developed a predictive model using the city’s public data over 10 years, including the history of orders, the number of children having lunch in the canteen and the composition of the menus. All this information was analyzed to develop a first model that lowered the overproduction rate by three points.
430 excess meals avoided per day
“By simulating 2018-2019 with our prediction model, we go from 9.6% of over-prepared meals to 6.5%, or 430 over-meals avoided per day“, explains Franckie Trichet. But he adds that”all is not won“. It is now necessary that the business teams responsible for orders appropriate this algorithm via a future tool – such as a website, an application or a dashboard – which remains to be found.
One thing is certain: this tool is not intended to replace these jobs but to help them in their daily decision-making. It should be operational by June 2021, estimates the deputy mayor. Clément Guillon, co-founder and COO of Verteegoo, adds that the algorithm will be further improved, which will allow it to expand its use cases. “For example, it can predict the number of students who go to lunch in the canteen and the impact of the menus on the number of children …“, he explains to The Digital Factory. In short, the predictive system is far from having expressed its full potential.
Brest and Montreal are interested
The initiative of the city of Nantes is innovative because it is part of an open innovation process. Indeed, the source code of the program is available on GitHub to allow other territories to take up this work and enrich it. According to Frankie Trichet, the cities of Brest and Montreal in Canada have shown particular interest in this approach.
–
