State of use of AI tax systems
The Kingdom of Denmark has always exhibited a high level of maturity in terms of automation and digitalization of society, including the use of machine-learning.
Prof. Brigitte Alfter reports in AlgorithmWatch that the Danish tax administration experimented with ‘traditional’ algorithms to automatically collect taxes (so-called EFI system – Et Fælles Inddrivelsessystem or One Shared Tax collection system) as early as 2005. The project was shelved in 2015 after major flaws with these systems were revealed.
According to reports from CIAT The Danish State pension system, SLP-S, also used algorithms to select taxpayers for control as early as 2001.
The Danish Tax Administration reportedly uses machine-learning since at least 2014.
What functions are performed with AI?
The machine-learning systems of the Danish Tax Administration perform a wide range of functions:
1. Webscraping: reports highlight that the Danish Administration makes use of open-source webscraping tools to automatically collect data from online websites, commercial platforms, social media and e-sharing/gig economy platform. Prior to open-source tools, sources indicate that the Danish Tax Administration leveraged XENON, similarly to Austria, the Netherlands, Sweden and the UK.
2. External risk-management (risk-scoring): the machine-learning systems segment taxpayers into categories of risks and select taxpayers with predicted high risk of fraud or non-compliance to devise an audit plan for the tax administration. The Danish Tax Administration uses supervised learning to attribute a score to taxpayers and determine precise treatment strategies in all areas of taxation, including VAT, personal income taxation, corporate income tax and customs.
3. Automated decision-making: the Danish tax administration uses an algorithm that automatically calculates and/or verify pricing of real estate without any individual manual inputs from tax officials. Reportedly, the model uses 19 variables, including proximity to facilities such as schools, parks, but also the level of pollution in the area. Algorithm Watch reports that the preparatory Committee in charge of the implementation of the ADM system considered machine learning to be more objective than human assessments.
4. Transaction Network Analysis: the Danish tax administration uses a real-time data matching system to flag suspicious VAT refund claims, since 2014. The system detect errors in reports and improper payments from the state Treasury. Reportedly, the system’s accuracy ranges around 70%. The system has now been updated and integrated to the Transaction Network Analysis system at the EU level (see EU Report).
5. Internal risk-management: the OECD ITTI reports that the Danish Tax Administrations makes use of AI to assist tax officials in making administrative decisions and recommendations for actions.
Reports of IOTA make incidental mentions of so-called ‘nudging tools‘, similarly to the Austrian, Belgian, and Polish tax administration. These systems adapt the language used on standard communication to taxpayers based on taxpayer historical data and underlying risk profiling. Concretely, it means that taxpayers may receive different types of communication based on underlying profiling and risk scores. For instance, notoriously non-compliant taxpayers might receive letters with a personalized ‘stronger’ language. Nudging tools may also be used to target specific segments of the population, to add specific text, such as the possibility to receive assistance from the administration for vulnerable populations.
What data can be processed by these systems?
The data used to train and processed by these different algorithms is not specified.
Reportedly the automated decision-making system used for real-estate pricing is based on a model of 19 features, including but not limited to, the sales prices of the real estate, its location, proximity to facilities and pollution in the area.
Are these systems regulated by specific norms?
The use of machine-learning by the Danish Tax Administration is not regulated by specific ad hoc norms.
The real-estate valuation system is regulated by the Ejendomsvurderingsloven (‘Property Valuation Act’).
