Poland’s machine learning STIR tackles missing trader VAT fraud
In 2017, Poland launched its System Teleinformatyczny Izby Rozliczeniowej (STIR) machine learing Artificial Intelligence (AI) programme. This uses algorithms to analyse large data of taxpayer activities to detect suspicious transactions, fraud and non-compliance.
This was in the backdrop of €6.6bn VAT Gap in 2017. Since then, the success of STIR and other compliance tightening measures has helped Poland cut its VAT Gap to €1.7bn by 2021 (latest figured), one of the most impressive rates of improvement in the EU. In addition to STIR, Poland also introduced JPK SAF-T VAT returns and many other measures.
STIR provides for the exchange of transactional and related data between various government authorities, notable the NRA and Central Register of Tax Data. This enables full analysis of daily transaction data provided by banks and credit unions.
Read more about how tax authorities adopt AI to raise tax revenues.
Machine learning of VAT data
STIR provides for three levels of this data analysis
- Risk scoring – based on analysis of the taxpayers trading history, where they are accessing bank accounts from, known movements of goods etc. This provides daily analysis in a data warehouse from the National Clearing House issued to NRA which can then opt to block bank accounts and launch investigations.
- Networking analysis – running reviews of links between taxpayers to help identify unusual and suspicious trades.
- Nudge communications – producing standardised reminders or alerts (letters, emails or texts) based on taxpayers’ activities to encourage them to undertake all compliance requirements.
Controversially, the algorithms used by STIR is not made publicly available. This has raised concerns particularly with many cases of innocent businesses having their bank accounts blocked. For example, the tracking of IP addressed used by taxpayers to log into their bank accounts.
Watch VATCalc’s new AI VAT Advisor deliver tax advice (legislative references and tax cases) on complex VAT questions.
