State of use of AI tax systems
Sweden’s Tax Administration (Skatterverket) has been a frontrunner in the use of machine-learning by EU tax administrations.
Earliest reports of the use of AI for tax compliance risk-management, date back to 2006.
What functions are performed with AI?
The Swedish tax administrations makes use of machine-learning to perform at least four functions:
- Risk detection: The Skatterverket makes use of an unsupervised learning model to detect under-reported income. It uses cluster analyses and nearest neighbours models to compare the incomes reported by taxpayers to other taxpayers in similar situations and detect outliers.
- Web-scraping: The Skatterverket makes uses of open-source webscraping tools to automatically collect data from online webpages, commercial platforms, social media, gambling websites and even on the deepweb. Prior to open-source tools, the Skatterverket made use of XENON, similarly to Denmark, the Netherlands and the UK.
- Social Network Analysis (SNA): the SNA algorithm visually represents a network of individual taxpayers as a combination of nodes and vertices. Using graph theory, SNA quantitatively and qualitatively measure connectivity between the nodes. Reportedly, the SNA model of the Swedish Tax Authorities was developed by SAS.
- Taxpayer assistance: The Skatterverket deployed ‘Skatti’, a chatbot able to answer taxpayers queries on personal taxation and population registration. According to Göran Sundin, digital strategist at the Swedish tax administration, Skatti handles around 15,000 chats per month and more than 200.000 queries since launch.
What data can be processed by these systems?
Skatterverket has not specified the data processed by these AI systems.
Are these systems regulated by specific norms?
The models are not regulated by specific legal norms.
On December 2023, the government submitted a proposal that would amend Swedish tax procedural norms to include new rules on the use of AI systems, such as risk-scoring, SNA and webscraping.
