The French public sector is not yet a champion of artificial intelligence. This was underlined by a report commissioned by the Prime Minister, which reported “under-exploitation by public authorities. »
The Ministry of Finance can no doubt act as a good student. In any case, the DGFIP’s activity report illustrates the interest shown in AI, particularly in terms of tax audits. In this sector, AI and data mining occupy a growing place.
Data-driven control activity
In 2022, 52.36% of checks were targeted using these techniques. This is almost 10 points more than a year earlier (44.85%) and 20 since 2020 (32.49%). This trajectory should continue. The DGFIP announces its desire “to extend the tools based on artificial intelligence. »
Control and sanction are not the only purposes in this area. Management also has the Weak Signals application. Thanks to this, it can “detect a company’s weaknesses early on. »
Signals’ algorithmic model was enhanced in 2022. It detected 20,509 companies in difficulty last year, of which 8,815 were analyzed and 4,164 selected for support.
Since 2022, the DGFIP has had a comparable tool, but this time focused on detecting local authorities in difficulty. But it is above all with its innovative land project that the administration has distinguished itself.
Swimming pool detection, a showcase for AI at the DGFIP
Designed with the help of technology partners, the detection model was put into operation in order to identify undeclared swimming pools. AI is used here to analyze aerial photos and detect taxable goods with a success rate of over 95%.
“A first experiment, relating to the detection of non-imposed swimming pools, gave positive results. The system has therefore been generalized to all metropolitan areas from September 2022”, indicates the report.
The generalization of Innovative Land resulted in the regularization by the tax authorities of around 125,000 swimming pools in 2022. The additional tax revenue for the State is estimated at 30 million euros per year.
AI is not the only use made by the finance department of its data assets. Since 2021, it has had a data lake, “significantly enriched” thereafter. Eventually, it will house all of the DGFIP’s data.
The Data Lake is enriched. And the use cases?
However, the administration still needs to develop uses and facilitate access to data. Improvements took the form in particular in 2022 of the use of data visualization. Through API, the DGFIP also wants to simplify exchanges and open up to the outside world.
“Requests for access to Public Finance APIs by external administrations are also increasing in number with 541 requests received in 2022 (bringing their number to 1,466 since 2019). »
The Digital Factory, an internal incubator for the development of digital services, could also contribute to a better use of data, and not only for the benefit of repression, for example in terms of immigration.
In its report on AI in the public sector, the Council of State recommended ensuring a balance of uses so that investments benefit agents and taxpayers. Otherwise ? The effects of “indispensable educational efforts” could be “annihilated by the feeling that AIS are first and foremost instruments of coercion. »