Zurich tests AI in the building permit process
Check building applications, check completeness, relieve authorities. What has long sounded like a dream of the future is now becoming a reality in the canton of Zurich. Based on a postulate from cantonal councillor Sonja Rueff-Frenkel, the cantonal government has commissioned two studies and has come to the conclusion that AI can help, but is not a panacea. From April 1, 2027, all planning applications must be submitted digitally. The starting signal for the next stage is now being given.
From April 2027, the canton of Zurich will require all municipalities to use the eBaugesucheZH platform. This lays the foundation for digitization, but only the foundation. The content of the applications will continue to be processed in different systems, depending on the municipality or canton. This historically evolved system landscape leads to media disruptions, manual coordination rounds and data inconsistencies.
What the FHNW study shows
The Building Directorate commissioned the FHNW Institute of Digital Construction to conduct a potential study along the entire process chain. 15 fields of action were identified, from initial digital information to building acceptance. The greatest short-term potential lies at the very beginning. Chatbots for the initial consultation, structured submission support and automated preliminary checks could immediately improve the quality of submitted applications and significantly reduce queries. Many improvements can already be achieved with rule-based systems, without generative AI.
Prototype with the city of Kloten
The Innovation Sandbox for AI of the Office of Economic Affairs tested an AI-based preliminary check for the notification procedure together with practice and technology partners, including the city of Kloten. For simple projects such as solar installations or heat pumps, a rule-based system automatically clarifies the admissibility and choice of procedure, and an AI then checks the completeness and quality of the entries. 3336 tests were evaluated. The results are encouraging, even if the reliable interpretation of complex plan representations remains an open challenge.
Humans remain responsible
Both studies agree that complete automation is currently not realistic. Where decision-making logic is clearly defined, rule-based systems are preferable to generative AI. The authority to make decisions remains with humans. Legal issues relating to data protection, liability, transparency and copyrighted blueprints as AI training material must be examined in depth before any implementation.
The results are now being incorporated into the further development of eBaugesucheZH. Individual applications are to be tested in pilot municipalities. Zurich is thus demonstrating how the careful, step-by-step use of AI can work in a complex administrative domain.