ChatAmsterdam is the result of a two-year partnership between the Municipality of Amsterdam and Aurai. During the first year and a half, Aurai’s consultant Don was closely involved in the technical development of ChatAmsterdam, focusing on building and refining the back-end infrastructure that powers the AI assistant.
In the final six months of the collaboration, the focus shifted from development to evaluation. This phase, led by Aurai consultant Bob, centered on researching how ChatAmsterdam was being used in practice by municipal employees. This case study is dedicated to that research.
The goal was to understand not only how effective the tool was in supporting daily tasks such as summarizing, drafting, and ideation, but also how it was being adopted by staff, what risks were emerging, and how responsible AI principles could be upheld in a real-world public service context.
The research was a collaboration between a multidisciplinary team, including DevOps Engineers, a Product Owner, and a Scrum Master. The team did not include a dedicated data scientist, hence why Bob was brought in for analysis using Python and Microsoft Azure services.
All chat data and prompts were pseudonymized and securely stored in Azure Blob Storage. Prompt classification was carried out using a large language model hosted on Azure AI Foundry, categorizing prompts by purpose, ranging from writing and summarizing to ideation and analysis.
Usage data was combined with qualitative insights from staff surveys, which captured real user experiences, expectations and concerns. The results were presented through interactive Power BI dashboards and regularly shared with project stakeholders.
Throughout the process, Aurai ensured strict adherence to the municipality’s data residency and compliance requirements. Security filters and governance frameworks were put in place to safeguard sensitive data and align with public values.
The results showed clear value in deploying ChatAmsterdam across the organization. Staff reported increased creativity, better focus on meaningful work, and a willingness to adopt new technologies. Repetitive tasks also became faster and more manageable according to the pilot users.
Perhaps most importantly, the project catalyzed a shift toward AI literacy: municipal employees gained confidence in experimenting with generative AI, supported by clear usage guidelines and trusted infrastructure.
Beyond efficiency gains, the project catalyzed a shift towards an AI-literate culture: employees felt empowered to experiment with new tools, supported by custom security filters and guidance that safeguard privacy, security, and digital autonomy. These carefully crafted mitigation measures ensure responsible use without stifling innovation.
The research concluded that ChatAmsterdam was ready for production use. A comprehensive report has been delivered to the municipality and is expected to be published on openresearch.amsterdam around Q3 2025.