Gemeente Hellendoorn
Dutch municipality Hellendoorn rolls out ChatGPT to staff and reports 2.54 hours saved a week
The Dutch municipality of Hellendoorn gave staff ChatGPT Business in a pilot that began at the end of 2024. In its own October 2025 survey, 116 of 164 users reported an average saving of 2.54 hours per week. The municipality scaled this to 11.57 full-time jobs.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
11.57
FTE
Time saved per week as extrapolated by the municipality from the survey average to 164 users1
78.3%
Share of respondents who said ChatGPT makes work more enjoyable1
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
Hellendoorn, a Dutch municipality, set an AI policy in early 2024. Its management team was keen to find uses. The municipality saw hurdles: working with AI takes different skills, the business case differs for each job, and it was unclear whether employees would want to use it at all. A licence alone, it concluded, would not be enough.
What they built
In late 2024 the municipality started a pilot that gave staff ChatGPT Business as a help in daily work. Alongside it, the project team set up ready-made assistants, which are custom versions of ChatGPT with fixed instructions and extra documents. The examples named in the slides include a writing-style helper, an assistant that follows the council proposal format, an HR assistant that answers questions from the collective labour agreement and staff handbook, and helpers for drafting social support plans. Others cover ICT procurement questions, grant decisions, a land policy assistant and, in testing, administrative law.
How they did it
Staff joined training by choice, in small groups, with function-specific examples. After two or three weeks the group evaluated, and each person said whether they wanted a licence. Short guides and an intranet group supported them. Use was voluntary and not tied to savings. By March 2026 there were 183 licences, and its slides describe ChatGPT Enterprise as the next plan, which they say keeps data in Europe and does not use it for training. Its slides also say that colleagues who are keen help others get started.
Results
A survey in October 2025 drew answers from 116 of 164 users. Of these, 78.3% said ChatGPT makes work more enjoyable, 77.4% said work quality improved, and 53.9% called it essential for their work. Respondents reported saving 292 hours a week in total, and 2.54 hours per user per week on average. The municipality scaled that average to all 164 users and arrived at 11.57 full-time jobs. Users said they spend the freed hours on work that would otherwise have stayed undone.
All reported figures
| Measure | Figure | Citation |
|---|---|---|
| Average time saved per user per week, self-reported in an October 2025 survey | 2.54 hours1 | |
| Time saved per week as extrapolated by the municipality from the survey average to 164 users | 11.57 FTE1 | |
| Time saved per week across the 116 survey respondents | 292 hours (8.11 FTE)1 | |
| Share of respondents who said ChatGPT makes work more enjoyable | 78.3%1 | |
| Share of respondents who said the quality of their work improved | 77.4%1 | |
| Share of respondents who said ChatGPT is essential for their work | 53.9%1 | |
| Survey response rate (116 of 164 users) | 71%1 |
Lessons for your project
Do not just hand out licences. The municipality found that a licence alone is not enough, because AI needs new skills and a different way of working.
Train in small, voluntary groups. Staff joined by choice, helped design solutions for their own roles, and decided after two or three weeks whether they wanted a licence.
Ship ready-made assistants. Pre-built assistants for council proposals, writing style and HR questions gave staff concrete starting points.
Expect team differences. The attitude of the line manager strongly affects adoption.
Say that AI is not a cut-back. The municipality made use voluntary and stated that it is not a savings measure.
Caveats
- All figures come from the municipality's own presentation of 12 March 2026, based on a staff survey. The sources do not describe an independent measure.
- Time savings are self-reported. The 11.57 FTE figure is an extrapolation by the municipality, not a measurement.
- The slides state the pilot started at the end of 2024 without a month; the deployed date shown is December 2024. They say 183 licences, while the survey base is 164 users.
- The slides give available budget as one FTE and 40,000, without a currency or an explanation. They do not state the pilot's total cost or the licence price.
- The pilot ran on ChatGPT Business. The slides name no model for it, and mention newer ChatGPT models only in the section on the Enterprise plan.
- The slides warn that no privacy-sensitive data should be entered in ChatGPT.
Sources
- AI binnen de Gemeente Hellendoorn: Tussen hype en werkelijkheid
Gemeente Hellendoorn (presentation to information managers) · 2026-03-12 · Conference talk · bouwstenen.nl
Summarized in our own words from the sources above. Spotted an error? See corrections.