Netcompany predicts AI savings potential of 90,000 FTEs in the public sector, but professors dismiss the analysis as unrealistic

Netcompany, the Danish IT company, says AI could free up 90,000 full-time equivalents (FTEs) in the public sector by 2035, worth around DKK 54 billion a year. The figure comes from the Netcompany Institute, its internal think tank, and goes well beyond the government's own target of 30,000 FTEs by 2035. Two-thirds of the potential would come from bespoke AI systems built into the specialised software and workflows of public authorities, and the rest from AI chatbots. Netcompany says standard assistants for caseworkers, teachers and social and healthcare workers would free only 11,000 to 17,000 FTEs once annual licence costs of DKK 0.9-1.9 billion are counted, so the government's goal cannot be reached by buying licences alone.

The method links the US O*NET database, covering 16,000 job roles, to Danish employment and wage data. Netcompany used the Claude Opus 5 language model to judge which tasks can be automated and how many minutes each would save, then added up the minutes. The company says no Danish equivalent of O*NET exists, and that time already freed by AI, for example among staff with an AI assistant, is not counted twice.

Two professors who study technology implementation call the assumptions unrealistic. Helene Friis Ratner of DTU Management says American job descriptions do not match how Danish welfare institutions organise work. In the US database, for instance, teachers' tasks exclude special needs education, which in Denmark is largely absorbed into inclusion and general classroom teaching. She adds that Danish teachers' preparation time has already been cut, so many of those gains are already realised. She and Ursula Plesner of CBS also say that breaking work into isolated tasks ignores how tasks interact, and that AI creates "invisible work" such as quality-assuring output, fixing errors and handling breakdowns.

Netcompany says it factored in time spent verifying AI output but did not calculate the cost of operating and maintaining AI systems, because those costs depend on specific solutions and management decisions. It assumes two working days of training per employee per year in total, which it calls realistic for standard chat assistants. Ratner calls that optimistic for a fundamental change in work processes.

The analysis also gives no cost estimate for developing the bespoke systems that account for most of the potential. Netcompany says a public-sector-wide estimate would need highly uncertain assumptions. The professors find this puzzling, given that large Danish public IT projects often run over budget and face delays.