State of use of AI tax systems
Publicly-available sources indicate that the Greek Independent Authority for Public Revenue (AADE) began using machine-learning systems around 2019, slightly later than its Western Europe homologues.
Since then, the AADE has operated a steady deployment of AI systems across all areas of taxation, most notably through their dedicated AI development centre.
What functions are performed with AI?
- Webscraping: Similarly to the French tax administration (see Country Report), the AADE leverages a scraping tool which collects data from private companies, such as e-commerce and social media platforms. The AADE has not disclosed much details on their scraping system, such as the data collected, the sources of collection or the taxes concerned. Reportedly, the underlying goal is to detect unrecorded financial activity through platforms and the sharing/gig economy. In 2020, the scraping system detected 20,000 properties for rent on Airbnb not declared to the AADE.
- Real-time risk detection: the AADE has developed a real-time data matching tool to detect and prevent missing-trader fraud schemes. The deployment of this tool coincides with the obligation for taxpayers to transmit revenues data to the myDATA platform in real-time. The tool also processes retail sales data of electronic cash registers.
- External risk-management (risk-scoring): the AADE deployed a supervised learning risk-scoring system to segment taxpayers into risk categories. The risk-scoring system generates a model that scores taxpayers to devise their annual treatment strategies and audit plans. The system is used across a wide array of tax items, including debt management prioritization.
What data can be processed by these systems?
The data processed by these AI systems is not detailed by the AADE.
Are these systems regulated by specific norms?
The use of AI by the AADE is not subject to specific ad hoc norms.
