How tax authorities use artificial intelligence to fight fraud

The tax administration is increasing the use of artificial intelligence every year to fight fraud and improve the efficiency of its services. In 2021, the share of checks noticed by AI was 45%, up from 32.5% a year earlier.

Over the past few weeks, teams from the Directorate General of Public Finance (DGFIP) have been introducing themselves to a new tool. Its name: Galaxy. Approved by the National Commission for Computing and Liberty (Cnil) on March 11, the new computer software will allow tax officials to better track fraudsters using automated processing of taxpayer data.

Its deployment on the premises of DGFIP is part of the PILAT project launched in 2018 to modernize the tax control system. The March 11 decree describes the new weapon as a “visualization tool” in the fight against fraud. Relevant elements of the financial and tax situation “.

Galaxy thus makes it possible to collect data such as reference tax income, spouse identity or even tax liability for individuals. However siren numbers, legal status, turnover class as well as taxes and duties which are subject to legal persons.

The adoption of this new tool by the tax administration marks an additional step by DGFIP’s desire to strengthen its digital arsenal in the fight against fraud. More broadly, the Barclays-affiliated service set up a Request and Evaluation Mission (MRV) a few years ago, with funding of 5.2 million euros over the period 2018-2022. Objectives: To develop artificial intelligence and data mining techniques in tax data processing and better target control and detection of fraud profiles. Recently, an algorithm has enabled tax authorities to mix and collect information from Internet users’ publications on social networks before processing them on a case-by-case basis.

45% of checks targeted by artificial intelligence

DGFIP is satisfied with the results of this 2.0 shift. In 2021, 13.4 billion euros were reported to individuals and businesses and 10.7 billion euros were collected. A level close to the 2019 record (11 billion) that the tax administration largely blames on “using data mining to better target tax audits”, where the share of audits targeted by artificial intelligence and data mining was about 45% last year, 2020- Compared to 32.5% and 21.95% in 2019. For 2022, it expects 50% targeted checks for AI

The tax authorities do not hide their intention to “increase the use of artificial intelligence”. And not just to fight fraud. Since March 2021, the deployment of digital assistants has made it possible to process the accounting posting of bank transfers obtained automatically in the collection of fines, and thanks to agents for improving the results of forced collections to increase the severity of the case. DGFIP.

Also in 2021, the tax administration began using artificial intelligence “to help contact center agents respond to user emails about payment issues”. It has also implemented the “Weak Signals” project, which is based on an algorithm “targeting companies’ vulnerabilities to set up support actions as soon as possible”. The system has made it possible for 23,227 companies to identify problems. Of these, 11,952 were selected for communication and assistance was offered.

Identify the swimming pool

In the experimental phase, DGFIP’s “Innovative Land” project aims to automatically detect buildings and swimming pools using artificial intelligence that have not been announced from the aerial view. The problem is, at this stage the device tested in nine sections with a 30% error rate will not really be created.

The tool will identify, for example, taxable swimming pools when they are above ground swimming pools are not taxable. Unions also report situations where artificial intelligence “pays taxes on buildings instead of tarpaulins, roads, sidewalks, parking lots, slabs”. Parisian Frederick Scalbert, CGT representative.

Unions mixed on the results of artificial intelligence in the fight against tax evasion. A report by Attac and Solidaires, supported by CGT, states that if this “material is interesting”, it “has not yet yielded all the expected results. Jobs, arguing that” digital will do better … “. , DGFIP staff was 94,669 in 2021.

Leave a Comment