State of use of AI tax systems

The first publicly available mentions of tax machine-learning algorithms used by the tax administration of the Republic of Latvia date to 2018.

What functions are performed with AI?

Based on publicly available data, the two functions performed by tax machine-learning algorithms in Latvia is:

  1. Taxpayer assistance: Latvia uses a VCA – chatbot named ‘Tom’, since January 2020 Tom answers frequently asked questions in particular in five tax relevant domains namely, registration of a business, e-filing, annual submission of tax returns, eligible expenses and the eligible expenses mobile application.

  2. External risk-management (risk-scoring): A report of the EU Regional Cooperation Council indicates that Latvia uses ESCORT, a machine-learning algorithm which ranks companies, rather than individual taxpayers, according to their predicted level of risk of fraud. ESCORT is particularly use to detect and prevent fraudulent or undeclared labor.

What data can be processed by these systems?

The data used for the development and use of the VCA is not specified. Reportedly, it mines the questions asked and answers provided by the VCA to update the quality of the service provided.

Regarding ESCORT, the different risk variables include: the average salary, the hourly wage the number of employees, the number of cash registers, the number of market places, the sector of activity, the number of offences detected during controls and the number of complaints received about envelope wages.

Are these systems regulated by specific norms?

The use of this model is not regulated by specific legal norms.