With the EU set to
implement mandatory e-invoicing and digital reporting by 2030, automated fraud
detection tools are expected to become standard in VAT enforcement. This
article seeks to address the potential risks of automatic fraud detection tools
implemented by EU Member States or at the EU level. The article examines these
risks and identifies four key concerns: the risk of treating algorithmic
predictions as evidence of fraud; limitations in data quality and model accuracy;
the potential for bias, and function creep; and the lack of legal protection.
These risks could have negative consequences for businesses. For this reason,
the article argues that such risks should be addressed through well-founded
decisions about the choice of analytical methods, based on the purpose of the
analysis, along with measures to prevent, correct, or compensate for
irregularities and discriminatory outcomes. Only sound and scientific
statistical research should be used, and systems should be regularly validated
and audited. Transparency in the operation of fraud detection tools is also
crucial. Most importantly, fraud detection systems should support human
decision-making, not replace it.