STIR — Poland's Real-Time AI System for VAT Fraud Detection

Introduced by a November 2017 law and operational since 2018, Poland's System Teleinformatyczny Izby Rozliczeniowej (STIR) is a clearing-house-mediated pipeline: banks and credit unions report account and transaction data daily to a Central Register of Tax Data, against which the tax administration's machine-learning risk-scoring algorithms run continuously to flag VAT carousel fraud routed through shell companies. Where manual analysis of such fraud previously took roughly two months, STIR enables detection in near-real time, and gives KAS the power to administratively freeze a suspected account for 72 hours, extendable to three months by court order, without a prior criminal proceeding.
Poland credits STIR, alongside e-invoicing and other compliance measures, with narrowing its VAT Gap from roughly €6.6 billion in 2017 to about €1.7 billion by 2021, one of the largest reductions of any EU member state over that period.
The system has drawn sustained criticism: the scoring algorithm itself is not publicly disclosed, and there are documented cases of legitimate businesses having accounts frozen based on opaque risk indicators. A peer-reviewed Intertax article examines STIR's compliance with privacy-rights standards, and AlgorithmWatch has documented affected businesses' experiences in an investigative report.

Read the full analysis: https://www.gov.pl/web/kas