State of use of AI tax systems

The first publicly available mentions of the use of tax machine-learning algorithms by the tax administration of the Republic of Poland (NRA) date back to 2017, with the adoption of the STIR system with the STIR legislation.

What functions are performed with AI?

Based on publicly available data, three functions performed with machine-learning were identified:

  1. Social Network Analysis (SNA) ‘ARANEUM’: ARANEUM visually represents a network of individual taxpayers using graph theory, it represents a network of taxpayers as a combination of nodes for individuals or points of interests and lines which quantitatively and qualitatively measure relations between the nodes.

  2. External risk-management (risk-scoring) ‘STIR’: System Teleinformatyczny Izby Rozliczeniowej or STIR attributes a score to taxpayers on the basis of risk-indicators determined by the central ‘clearinghouse’, following analyses of taxpayer historical data. STIR is used by the Polish tax administration regarding the VAT tax burden of financial institutions and enables the real-time exchange of data between financial institutions, the NRA and the central data warehouse (Central Register of Tax Data). Risk-scores are transmitted daily to the NRA who can then decide to block taxpayers’ bank account for 72 hours (so-called freezing). This period can be extended to 3 months in case of serious suspicions of fraud (extended freezing). M. Rojszczak reports that in the period of January to June 2019, 192 accounts linked to 39 entities were blocked which amounted to approximately 5.5 million euro gains (0.007% of the total VAT gap for that year). A. Bal reports that in 2018, 30.000 accounts were flagged, but only 23 entrepreneurs had their account blocked.

Incidental reports of IOTA officials mention the fact that the NRA has also been experimenting with nudging tools, which adapt the language of standard communication to taxpayers, based on profiling of taxpayer historical data, to nudge these taxpayers to compliance without using more coercive means.

What data can be processed by these systems?

The data used by ARANEUM and by nudging tools is not specified. Regarding risk-scoring algorithms, STIR captures in real-time and remits daily to the NRA data provided by financial and banking institutions including data on bank accounts, daily history of transactions, identification data of senders and recipients of goods and services, initial and final balances statements, IP adresses from which account holders logs into their accounts.

Are these systems regulated by specific norms?

ARANEUM and nudging tools are not regulated by specific legal norms. The STIR system is regulated by the STIR legislation.



  • System Teleinformatyczny Izby Rozliczeniowej – Tekst jedn. Dz.U. z 2017 r., poz. 1221 ze zm, 3 (4).


  • A. Juhasz, ‘IOTA Webinar on Social Network Analysis’ – presentation of Krzysztof Śledzikowski ARANEUM – the use of SNA techniques and graph databases to VAT fraud detection’ (March 2021) – available at: - last accessed July 2022. ​
  • A. Bal, ‘Ruled by Algorithms: The Use of ‘Black Box’ Models in Tax Law’ [2019] Tax Notes International, Vol. 95, N°12, pp. 1159-1165; see also A. Bal for VERTEXINC – available at: - last accessed July 2022. ​
  • M. Papis-Almansa, ‘The Polish Clearing House System: A ‘Stir’ring Example of the Use of New Technologies in Ensuring VAT Compliance in Poland and Selected Challenges’ in [2019] EC Tax Review, Vol. 28, Issue 1, pp. 43-56
  • M. Rojszczak, ‘Compliance of Automatic Tax Fraud Detection Systems with the Right to Privacy Standards Based on the Polish Experience of the STIR System’ [2021], INTERTAX, Vol. 49, Issue 1, pp. 39-52
  • M. Macudzinski, ‘System Teleinformatyczny Izby Rozliczeniowej – nowe narzędzie zwalczania wyłudzeń skarbowych’, Prawo Budzetowe, Panstwa, Samorzadu – DOI: - last accessed July 2022. ​
  • Konrad Szczygiel, ‘Pre-Crime at the Tax Office: How Poland automated the fight against VAT fraud’ in AlgorithWatchm – available at: - last accessed July 2022. 
  • Hernandez et al., Applying Behavioural Insights to Improve Tax Collection: Experimental Evidence from Poland (World Bank, 2017).