evidoria

← Back to browse

Good practice Imported

ChatAmsterdam — Amsterdam's In-House Generative-AI Assistant for Municipal Civil Servants

Netherlands · Amsterdam · See the Netherlands profile · See the Amsterdam profile

Evidence: Observational / pre–post Top 62% 53/100 · Ask Evidence Copilot about this practice

After tightening its generative-AI rules in February 2025, Amsterdam piloted ChatAmsterdam with 800+ civil servants over six months (34,000+ prompts, 10,000+ chats), then launched it municipality-wide in May 2025 as a data-sovereign alternative to commercial chatbots.

800+ civil servants
Pilot participants (Nov 2024-May 2025)
34,000+ prompts
Prompts logged during pilot (Nov 2024-May 2025)
10,000+ chats
Chats logged during pilot (Nov 2024-May 2025)

Details

Maturity
Established
Promoter
Gemeente Amsterdam (City of Amsterdam)
Period
2024–2025
Region (NUTS)
NL32
Keywords
digital government, public sector productivity, generative AI, internal operations, data protection

Context

In February 2025 the City of Amsterdam tightened its generative-AI policy, restricting staff use of ChatGPT, Microsoft Copilot and Midjourney over data-protection concerns, and committed to building an in-house alternative.

Objectives

The city set out to give civil servants a data-sovereign chatbot for drafting text, producing summaries and analysing complex documents, without sending municipal data to external commercial models.

Activities

Amsterdam ran a closed pilot from November 2024 to May 2025 in which more than 800 civil servants tested ChatAmsterdam, built on GPT-4 but configured so entered data stays within the municipality's own environment and is not used to retrain the underlying model.

Results

The pilot logged more than 34,000 prompts and 10,000 chats, and on 8 May 2025 Amsterdam rolled ChatAmsterdam out to all municipal civil servants.

Conclusions

No independently measured productivity or accuracy figures from the pilot have been published, and the reported time-saving expectation is self-reported rather than verified against a baseline; Amsterdam intends to share its approach with other Dutch municipalities and connect future European-built language models as they mature.

Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Medium (1–3 years)
Staffing & skills
City of Amsterdam in-house build/policy team, 800+ pilot civil-servant users

Conditions for success

  • a prior municipal policy tightening (Feb 2025) restricting external commercial AI tools, creating demand for a data-sovereign alternative
  • configuring the assistant so entered data stays within the municipality's environment and is not used to retrain the underlying model

Common failure modes

  • no independently measured productivity or accuracy outcomes have been published
  • time-saving benefit is self-reported rather than verified against a baseline

Commonly funded by

National / regional programmes Digital Europe Programme

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful