Artificial Intelligence at the Finnish Tax Administration

AI can help us reach our strategic goals: ensuring tax revenue, fair tax assessment, and positive taxpayer experience.

Why do we use artificial intelligence?

By harnessing AI, we can provide better service for our customers and operate more efficiently.
While AI takes care of some of the work, tax officials can concentrate on cases that require human expertise.
AI makes tax assessment quicker and more efficient, and it reduces the risk of human error.

Our use of AI is governed by legislation and other regulations concerning our operations, such as

the principles of good governance, civil service ethics, and the values of public administration
the principles of tax assessment

We want to be responsible and ethical in the ways we use AI. To this end, we have compiled the Finnish Tax Administration’s ethical principles for AI, which guide our in-house use and development of AI. These principles are as follows:

We use both off-the-shelf AI solutions and solutions developed in-house.

AI can offer support and help to tax officials. AI can never make decisions independently. The final decisions must always be made by a human.

Scope: Tax assessment
Starting point: Every year, the Tax Administration receives 2.8 million pages of tax documents on paper. All documents received in a non-electronic format must be digitised so that they can be processed for the purposes of tax assessment.
AI solution: The Tax Administration's service provider uses a rule-based AI solution for digitisation.
- The service provider performs pre-processing on the forms, including opening the envelopes and removing materials that cannot be scanned, such as memory sticks.
- After this, the documents are scanned as images and the data content is read using optical character recognition (OCR). The checking process of the data retrieved utilises machine learning. In uncertain situations, the system will highlight data elements that need to be checked by a human.
- Finally, the files are transferred to the Tax Administration's taxation system for processing.
Benefit: Using a rule-based AI solution speeds up the digitisation of data and reduces the amount of human work needed for the process.

Scope: Customer service
Starting point: The Tax Administration switchboard receives a large number of calls that must be manually forwarded to the service numbers for tax cards and individual taxation.
AI solution: In collaboration with the operator, a speech recognition bot has been implemented to the switchboard.
- When a customer calls the switchboard, the bot asks what they are calling about. The customer briefly describes the matter.
- The bot then forwards the phone call either directly to the service number for tax cards or individual taxation or to the human-operated switchboard.
Benefit: The customers can access the services they need faster and we save resources in call forwarding, when the AI directs calls automatically to the correct service number.

Scope: Customer service
Starting point: When we started offering a chat service, the demand was so high that the human resources available were not enough to meet it. All service agents were constantly busy and the queue was full, so the chat service was unavailable for most customers. In addition, taxpayers need help with using MyTax and taking care of their tax matters outside service hours, too.
AI solution: We have implemented a chatbot in MyTax and on vero.fi that uses Tax Administration AI to answer easy and frequently asked questions.
- The chatbot uses natural language processing methods in order to respond to customer questions in real time.
- Our chatbot solution has been acquired as an off-the-shelf service, but we manage the chatbot independently and in-house. For example, we provide the answers the chatbot can give, we take care of its training, and we monitor its functioning.
Benefit: The chatbot allows us to meet the demand for chat service. In addition, taxpayers can get help in using MyTax and other such issues regardless of the time of day. This allows for a greater number of taxpayers to take care of their matters independently.

Scope: Communication
Starting point: The Accessibility Directive requires that the videos and sound files used in customer communication are subtitled. However, producing subtitles manually is time-consuming and expensive.
AI solution: The Tax Administration uses an AI solution for generating subtitles for video and audio files. We have developed the solution ourselves in-house.
- The video or audio file is uploaded to the application and the application generates a text file based on the input file. The text file can also be edited as needed. The text file can then be attached back to the video or it can be directly utilised in text format.
- Our text generation application utilises a number of cloud services. The application has been developed at the Tax Administration and is intended only for use in the Tax Administration's communications.
Benefit: Previously, subtitles were purchased from a third party. The annual costs of the AI solution are significantly lower than those of the purchased service.

Scope: Communication
Starting point: The vero.fi website is continuously developed to meet the evolving needs of taxpayers and to match the general development of the sector. The site regurarly needs updated illustrations. These illustrations must take accessibility, regional equality and other such concerns into account. Organising photoshoots is time-consuming and expensive.
AI solution: The Tax Administration has collaborated with an artist specialising in AI imagery to generate AI illustrations for the vero.fi website.
- The images portray environments that could be located anywhere in Finland. The images also change according to the seasons.
- The Tax Administration has complete rights to use the illustrations and to continue developing the imagery further with AI solutions.
Benefit: AI is a cost-effective and quick solution for producing imagery that can be used on vero.fi.

AI refers here to both rule-based solutions and systems utilising statistical machine learning.

Page last updated 6/5/2024