A thousand AI pilots in Swedish municipalities. How many become everyday work?

A lot is happening with AI in Swedish municipalities, which is encouraging. But when I talk to department heads, I often hear the same thing: we ran a few trials, they looked promising, and then not much more happened.

The breadth is already there

When AI Sweden mapped municipal AI work in 2024, it found just over a thousand AI-related initiatives in 263 of the country's 290 municipalities. At the same time, it noted that many are working on the same thing and that there is a risk of initiatives getting stuck at pilot stage. A 2025 survey by the trade union Akavia points the same way: every municipality that responded uses AI in some form, but use is still limited and unevenly spread.

So interest is not the problem. What is missing is the path from trial to way of working.

What I see in many organisations is a fear of AI. It often comes from within the organisation itself, from the very functions that ought to be driving the issue but instead hold back and point to the risks. I believe it comes down to a lack of knowledge, and sometimes a lack of courage. The risks are real, but they can be managed with the right knowledge. Those who never dare to try, on the other hand, never learn which risks are real and which are just worry.

Why pilots stall

The pilot has no recipient. A trial run by a digitalisation strategist without a service manager who wants to take it over has nowhere to go when the trial ends.

Nobody has calculated the benefit. It is hard to get money to scale something up if nobody can show what it would deliver.

Legal and information security come in too late. When questions about personal data and public access to records surface after the pilot, they become a blocker instead of part of the design.

How to improve the odds

  1. Only start pilots that have a business owner who has said she wants to take it over if it works.
  2. Decide before you start what must be true to go further, and what would make you stop.
  3. Bring in legal and information security from day one. It makes things faster, not slower.
  4. Copy rather than invent. If another municipality has already solved the same problem, start there.

A question for your next leadership meeting

Which of our AI trials has a service manager who wants to take it over when the trial ends?

Sources: AI Sweden, mapping of municipal AI work, 2024 (Swedish) · Akavia, AI in municipal service, 2025 (Swedish)

What does it look like in your organisation?

Take the assessment on AI in municipal development to see where you stand, or book 30 minutes with me about the next step.