Federal Tax Administration (ESTV) relies on AI for its knowledge management.
With the Proof of Concept “PocKnowAI,” a generative AI based on Microsoft Azure AI was successfully implemented. It delivers precise, context-relevant answers with transparent source citations.
Initial Situation
The ESTV has an extensive but unstructured knowledge base in the field of tax law. This knowledge exists in various systems and formats — such as PDF documents, HTML pages, or XML documents.
To answer taxpayers’ tax law inquiries, ESTV employees had to spend considerable time searching for and extracting relevant information from these knowledge pools. This challenge was further compounded by a demographic-driven loss of expertise.
Solution
As part of a Proof of Concept, a generative AI was successfully implemented to support ESTV employees. The system simplifies access to publicly available specialist knowledge by providing precise, context-relevant answers. Particular attention was paid to transparent and accurate source citations, with a linking system realized at the article and page level.
The platform was enriched with a limited number of specific documents from the areas of value-added tax and withholding tax, as well as relevant legal texts, regulations, and court rulings.
Thanks to the iterative approach and continuous improvements, the hit rate for document searches increased significantly over the course of the project.
The AI platform is based on Microsoft Azure AI, with the user-friendly interface built on Microsoft Azure and Microsoft .Net.

























































