State of use of AI tax systems
As of October 2024, the Croatian Tax Administration (CTA) has not publicly disclosed the integration of machine learning and AI systems.
The Croatian tax administrations reported not to use artificial intelligence both in the International Survey on Revenue Administration (ISORA) and the OECD Inventory of Tax Technology Initiatives.
Nonetheless, systematic review of available literature shows that the Croatian tax administration has integrated a predictive tool to detect and prevent carousel fraud, connected with the Transaction Network Analysis network (see EU Report).
What functions are performed with AI systems?
Available literature highlights that the Croatian tax administration makes use of machine learning to perform two functions:
- Risk detection: CTA has integrated a tool developed by CROZ to enhance existing detection processes. The system infers features from taxpayer historical data to predict future risks of non-compliance. CROZ indicates that model development was supported by EU funds.
- Transaction Network Analysis: CROZ indicates that the risk detection tool is equipped with network analytics to quantify the connectivity between taxpayers and prevent carousel schemes.
What data is processed by these systems?
The data processed by the AI system was not specified by the CTA.
Are these AI systems regulated by specific norms?
The use of AI by the Croatian tax administration is not subject to specific ad hoc legislation.
