Automated artificial intelligence and machine learning supported tax registration for non-Swedish customers

Countries

Sweden

Policy areas

Organisation name Swedish Tax Agency

Contact person: Björn Erling

bjorn.erling@skatteverket.se

In 2021, Sweden introduced new legislation extending Swedish tax liabilities to temporary workers having no permanent residency for tax purposes. The registration process that was already in place for non-Swedish workers was manual and paper-based, and peaks in volumes often created delays, sometimes for several weeks. Furthermore, it was not compatible with the tax registration system for Swedish citizens, which is digital, fast and seamless.

Non-Swedish residents needed to get the correct tax registration, preferably before commencing work or business activities in Sweden, to proceed with their commitments without incurring potentially unknown tax risks.

International taxation rules are complicated, and normally new, non-permanent residents had to provide a lot of personal and work-related information for correct taxation. At the same time, identification and risk assessment of foreign residents, highly relevant when it comes to fighting fraud, were very difficult tasks due to the use of temporary non-Swedish workers in labour-intensive sectors and the complexity of international tax and treaty rules.

The process needed to be smoother and meet the requirements of EU law regarding free movement, and to fulfil internal and international obligations to reduce risk.

To tackle this, the Swedish Tax Administration (STA) created the TAIS project (Tillfälligt Arbete i Sverige [Temporary work in Sweden rules]. It consisted of the implementation of an automated artificial intelligence and machine learning supported tax registration for non-Swedish residents, which made the process faster, less liable to errors and fraud-resistant.

The first phase started in 2019 and the service was opened to non-Swedish residents (for both individuals and businesses) by summer 2020. It implied digital applications and enabled training of the risk assessment models integrated with case management.

The service was set up within the Swedish Tax Agency technical environment and is currently managed by teams with Agile Release Trains, which guarantee the further improvement and sustainability of the project.

The possibility to handle all incoming matters manually was introduced immediately. However, the manual case management remains, to train a new ML-based (AI) learning model for risk management, and to implement an AI/algorithm-based support service. This is continuously going forward to help manage non-standardised information (e.g. foreign government documents and ID documents) for automatic interpretation and analysis.

The intention is to use the results of the AI-based risk management to allow more temporary residents to register in a completely automated manner, close to real time. This is despite the information required being partly unstructured documentation.

The STA has estimated that the AI will reach good levels of accuracy in early 2022.

Other countries manage the same type of semi-structured information in various types of application procedures for foreign residents, and need to build capacity around processing that type of information for automated outcomes. This is even more useful, as long as all documentation taken from countries of origin and presented to the receiving states are not internationally standardised and digitised worldwide. This is now possible by using machine learning and AI-based technologies.

Also interesting