The Czech Financial Administration is testing artificial intelligence designed to make life easier for taxpayers and auditors. The application can pre-fill a tax return from uploaded documents, and analytical tools help process hours-long interviews. Decision-making, however, remains with people, and the projects are moving forward through pilots, with attention to security and legislation.
AI that pre-fills the tax return
A citizen uploads photos of certificates from employers or documentation for deductions, such as for a mortgage or blood donation. The AI extracts data from the documents, sorts it by type (income, deduction, etc.), and pre-fills the return, which the user reviews and then submits in the tax system. Responsibility remains with the taxpayer and verification is essential, but the manual search for the right fields is eliminated.
The application is not connected to registers and works as a voluntary aid. It has gone through several rounds of testing with employees, students, and professional chambers, and a public pilot is being prepared. Looking ahead, integration into the Moje daně portal is being considered to make use even simpler. The aim is also to help in situations where employers are increasingly not preparing the return on behalf of employees.
Faster processing of interviews for auditors
Another project focuses on recording, transcription, and analysis of interviews in tax audits. The tool extracts important entities and key moments from the transcript and allows them to be linked across multiple case files. It saves analysts time when working with large volumes of material, but the final assessment and decision remains with the person who was present at the proceedings.
Technically, it is an on-premises solution that requires robust infrastructure and clear choices: should the transcript be verbatim, or is semantic accuracy sufficient? Process in real time, or overnight, which affects capacity? Models are evolving quickly, so an agile approach and the ability to change technologies flexibly without impacting the entire system are important.
People, rules, and gradual evolution
Although the projects are not in production, interest in the pilots was surprisingly high across age groups. Employees recorded dozens of hours of mock interviews with dialects, overlapping speech, and background noise to enable thorough testing and tuning of the models. Clear rules about where to use AI and where not to, along with voluntary use, are key: the new tools are an aid, not an obligation, and the traditional process remains available.
A few recommendations were voiced for Slovakia: build on experience with chatbots built on top of guidelines, build shared infrastructure and platform services for rapid application development. Alongside the technology, lead an open dialogue with universities and the public about legal and ethical boundaries so that trust builds gradually. According to Czech experience, people in Slovakia often embrace innovation more readily, which can accelerate the sensible, evolutionary adoption of useful tools for taxpayers and approximately 14,000 tax administration employees.
