How Bercy uses artificial intelligence to target its tax audits

New technologies increase the results of tax audits. The technological development makes it possible for the tax administration to make decisive progress.

Bercy has prioritized the search for information in connection with the fight against fraud. It is a matter of revealing the mechanisms of fraud, collection, centralization and enrichment of external information and information dispersed in the services to suggest commitment to tax audit.

The benefits of data mining

The tax administration’s team’s work with the databases makes it possible to increase the efficiency of tax audits. The use of databases (data mining) is optimized through the use of artificial intelligence. In 2021, almost one in two checks were carried out after an analysis of data mining. According to the latest report of the DGFiP published in May 2022, the productions from the data project Targeting fraud and valuation of requests (CFVR) at the end of December 2021 represented more than 44% of the programming of tax audits, with a target of 50% in 2022. CFVR is a large-scale project developed in 2014, which aims to improve the efficiency of tax audit operations and is part of the modernization of data analysis work and targeting. Funded by the Public Action Transformation Fund, it explores information from 11 databases, such as Sirius-Pro, the tax audit software of the Directorate-General for Public Finance (DGFiP) or the official bulletin for civil notices and commercial (Bodacc). Relying on exploratory methods based on statistics and algorithms and by making it possible to model behaviour, this tool uses artificial intelligence techniques to automatically identify cases of possible tax fraud.

A dedicated incubator

To innovate further, the tax administration relies on an internal incubator. With the Digital Factory, the delegation for digital transformation created at the beginning of 2021 an experimentation laboratory, the Digital Factory, an incubator dedicated to DGFiP agents who want to develop new digital public services. In 2021, it made it possible to develop two new projects: a series of tools to support small communities in the establishment of their investment projects and a new application to facilitate the collection of the tourist tax by making reliable and collecting data from the rental of the furnished tourist accommodation , specifies the latest activity report of the General Directorate of Public Finances in June 2022. In fact, data or data and their mass exploitation thanks to artificial intelligence is now one of the main priorities of Bercy.

Building a data lake

The constitution of a data lake was initiated in 2021. It will eventually include all data from the DGFiP, in order to optimize their use thanks to easy, centralized and secure access. The aim of this project is to gradually bring, and depending on the pilot projects, a good part of the available data from DGFiP into an infrastructure dedicated to the processing of large amounts of data, which is accompanied by services and additional tools dedicated to data science, visualization, or query tools, allowing users to better appreciate the data available in the data lake. The data lake is therefore a Big Data processing tool. The initial work on property transactions made it possible to identify undervalued or overvalued transactions and recall more than 2.7 million euros. The Innovative Land project is developed within this framework.

The experimentation of the Innovative Land project

In order to guarantee a better reliability of the basis of local direct taxation, DGFiP within the framework of this project uses innovative technologies for artificial intelligence and valorization of data from aerial photographs from the National Institute of Geographical and Forest Information (IGN) . It aims to detect in these aerial photographs, swimming pools and other black buildings, when they should be directly levied on local taxes. Specifically, the algorithms make it possible to extract public aerial images from IGN, which can be consulted by anyone on the website, the contours of buildings and swimming pools. A computer processing then verifies, based in particular on the declarations of the owners to the city planning and tax administration, whether the elements thus detected in the images are correctly subject to local direct taxes. A tax management agent then systematically checks each detected irregularity before any follow-up by the property owner and ultimately of taxation. An experiment was launched in October 2021 in nine departments in the South-East (Alpes-Maritimes, Var, Bouches-du-Rhône, Ardèche, Rhône, Haute-Savoie) and West (Morbihan, Maine-et-Loire and Vendée). with a primary focus on the detection of swimming pools. The operation will then be extended to include black or wrongly taxed buildings. This project aims, by optimizing the process of detecting black constructions or developments, to make it possible to fight declarative anomalies more effectively and thus better meet the citizens’ wishes for fairness and fiscal justice through fair taxation of goods. This project will also make it possible to improve the reliability of the basis of local direct taxation and to facilitate the representation on the cadastral map of buildings and swimming pools from aerial views.

A new Galaxie application to optimize tax audits

The tax administration continues to develop new tools such as the Galaxie application, a computerized and automated processing of personal data. Third block of the PILAT project (PILotage and Analysis of Control), GALAXIE has been implemented since April 5, 2022. The PILAT project, which started in 2018, is a major project to modernize the information system dedicated to tax control. Its goal is to provide a single application for the entire tax audit chain, from programming to collection. The GALAXIE application was approved by a decree of the Minister of Economy (decree of March 11, 2022 on the authorization of the DGFiP for the processing of personal data called GALAXIE) following the positive opinion of the CNIL (consultation No. 2022-025 of February 17, 2022, issuing a opinion on a draft decree establishing by DGFiP the processing of personal data called GALAXIE). It is a visualization tool at national level, on the one hand, of the links between professional entities (links of participation) and between professional entities and natural persons (links of manager, partner or shareholder) and, on the other hand, elements of context about these persons’ financial and fiscal situation. For companies and legal persons or entities, these are data relating to the refund of VAT deductions, collective legal proceedings, fees paid, telephone number, e-mail address, indicator of sensitive persons, data relating to tax compliance, the administrative department of the tax file, data relating to to the bank account, managers and partners’ links. For natural persons, this concerns data relating to indicators of sensitive persons, the service managing the tax file and connections between managers and partners. This tool is intended for authorized agents, with the purpose of improving compliance with tax obligations, to conduct research, investigation, programming, control and recovery of tax violations. It will allow them to overlay and cross-check a lot of information, including the existing links between natural persons and professional entities, director, partner or shareholder links or between different entities, such as the links to participation, by putting them in contact with situations heritage and economy. This application will make it possible both to identify mistakes made by taxpayers in good faith and to facilitate the detection of tax evasion schemes and, in particular, to establish joint responsibility in the organization of fraud thanks to a union of tools, which are now divided into departments.

Increase the use of artificial intelligence

The tools developed provide a gain in productivity and efficiency by providing all actors in the tax audit chain with modern tools such as data visualization, automated and open to collection and litigation. It will also make it possible to strengthen and streamline the management of the activity in an integrated digital environment, to eliminate redundant tasks, while at the same time automating procedures. In general, DGFiP intends to increase the use of artificial intelligence. The use of digital assistants frees departments from time-consuming and repetitive tasks and allows agents to concentrate on missions where their expertise will be better valued. Thus, since March 2021, the deployment of digital assistants in the fine collection network has made it possible to automatically process the accounting entry of bank transfers received and for agents to improve the results of enforcement thanks to the intensification of prosecution. The results recorded since its commissioning are positive: the processing rate of wire transfers received by DGFiP is above 50% on average, with peaks that can exceed 90% of transfers received for certain accounting items. In 2021, DGFiP began using artificial intelligence to help contact center agents respond to user emails about payment issues. In this connection, it has initiated a modernization of the standard answer database and opening access to all DGFiP documentation via a search engine. So many first steps in a paradigm shift in tax auditing.

Leave a Comment