State of use of AI tax systems

According to a report of the Austrian Federal Ministry of Finance and literature from IOTA, the Austrian tax administration has been making use of machine-learning algorithms since at least 2016, when the Predictive Analysis Competence Center (PACC) was created, following the EU customs risks management and action plan of 2014.

What functions are performed with AI?

The report of the Federal Ministry of Finance is particularly vague about the specific functions performed by the machine-learning algorithms. Nonetheless, the report mentions two specific functions:

1. Risk-management systems (risk-scoring): The tool creates risk profiles to segment individual taxpayers into categories of risks and operate a centralised selection of taxpayers to be audited.

2. Real-time risk detection: the tool identifies suspicious transactions and flags those for potential further audits.

Other sources from IOTA, and in literature make incidental mentions of so-called ‘nudging tools‘, i.e. tools which adapt the language used on standard communication to taxpayers based on predicted risk-factors and taxpayer historical data

What data can be processed by these systems?

The report of the Federal Ministry of Finance mentions that the PACC has at its disposal all national and international taxpayer data available, which are stored in a central data warehouse.

Are these systems regulated by specific norms?

The models used are not regulated by specific legal norms, but were implemented based on internal policy decisions of the Federal Ministry of Finance, following the EU action plan of 2014.