State of use of AI tax systems

According to data from the OECD, the Finnish Tax Administration has developed and has been using two machine-learning algorithms since as early 2016.

What functions are performed with AI?

  1. External risk-management (risk-scoring): The tool assesses fiscal risks of tax processes carried out and segments individual taxpayers into categories of risks to subsequently prioritize the taxpayers to be selected for audits.

  2. Taxpayer assistance: Virtual conversational assistant ‘Virtanen’ – The chatbot Virtanen is used to assist taxable persons by providing automated guidance for natural persons on the MyTax platform. It can converse with taxable persons, and forward non-solved queries to human tax officials. During 2020, Virtanen had 318,974 conversation and only 18.1% of conversation – 57,873 conversations were forwarded to human agents, hence the chatbot had a rate of success of roughly 81.9%.

What data can be processed by these systems?

The risk-scoring tool analyses the data stored in the Incomes Information System (tulotietojärjestelmästä) and the Business Information System (BIS), two central data systems where all national and international tax information (wages, dividends, benefits, insurances, social contributions) are stored.

The data used to develop and used by the chatbot Virtanen is not specified in publicly available sources.

Are these systems regulated by specific norms?

There is no specific legal basis for the use of the two AI tools. According to the Finnish Tax Administration the use of these tools is guided by the principles of the tax assessment procedures and the Tax Administration Act (503/2010). The two tools were implemented on the basis of managerial decisions of the Finnish Tax Administration.

The use of these tools also follows the principles for ethical AI of the Finnish Tax Administration.