Poland Wields AI in the Battle Against VAT Fraud

Poland's Plan To Strengthen VAT

Value added tax (VAT) provides a big part of Poland’s revenue, but 10 years ago the country was losing much of that income to fraudsters. The VAT shortfall soared from 0.4% of GDP in 2006 to 1.5% in 2011, hitting an estimated USD 11.2 billion the following year. In response, Poland started implementing a comprehensive plan to strengthen its VAT system. In 2017, it unveiled System Teleinformatyczny Izby Rozliczeniowej (STIR), an innovative measure to tackle VAT fraud that employs an artificial intelligence (AI) agent to analyze VAT-relevant transactions.

How STIR Works

Here’s how it works. STIR enables the exchange of data among financial institutions, the National Revenue Administration (NRA) and the Central Register of Tax Data. Banks and credit unions report account data and clearinghouse data on a daily basis for all transactions carried out by entrepreneurs. The clearinghouse uses algorithms, based on criteria used by the financial sector, to calculate a risk indicator for each entrepreneur. The criteria may include, for example, customer residence, the complexity of the entrepreneur’s ownership structure and unusual transactional circumstances.

It’s not an entirely machine-based process, however. The clearinghouse sends the data to the NRA, and the head of that agency may decide that an entrepreneur is at high risk of being involved in VAT fraud. If so, the agency may take administrative measures such as blocking a bank account for up to 72 hours, or as long as three months in some cases.

STIR's Success

Poland can claim some successes from STIR. It enables the NRA to detect potential carousel frauds in near-real-time, versus the two months that might have been needed previously. In 2018, almost 30,000 entrepreneurs were flagged as high-risk, although only 23 had their accounts blocked.

Interestingly, STIR is a “black box” AI model, in the sense that the algorithms used to determine the risk indicator are not disclosed to taxpayers. The question of visibility into AI models is an increasingly important one for tax law, and I take a deeper look at the issues – including the arguments around STIR – in my new Tax Notes International article.