AI for public space reports: faster to the right department
Short answer
Public space reports are a classic example of AI that works: high volume, short texts and mistakes that are spotted quickly. Amsterdam has categorised reports for years with a machine learning model in Signalen, its open-source reporting system. Language models add a step: interpreting photos, summarising a report and sending the resident clear feedback.
An example that already runs
Amsterdam processes hundreds of thousands of public space reports a year. In Signalen, the reporting system the city makes available as open source, a model uses the words in a report to decide which category it belongs to and which department picks it up. It is described, with explanation, in the Dutch Algorithm Register.
That is exactly the kind of application to start with: high volume, a clear outcome, and a wrong category is quickly corrected. See Where to start with AI automation.
What a language model adds
| Step | Classic model | With a language model |
|---|---|---|
| Categorising | Based on words | Also for vague or long reports, and in more languages |
| Photos | No | A description of what the photo shows, for checking |
| Duplicates | By location | Also by content: the same tree, described differently |
| Feedback | Standard text | A clear message on what happens and when |
| Reporting | Counts | Summaries of recurring problems per neighbourhood |
What to watch for
- Personal data in reports. Residents mention names, number plates or neighbours’ addresses. Process reports in your own environment, not at an external AI service.
- Photos with people. A photo of a badly parked car shows a number plate; a street photo sometimes shows faces. Keep them no longer than necessary.
- Equal treatment of neighbourhoods. If prioritisation plays a role, check that areas with fewer reporters do not fall behind.
- The register. A model that routes or prioritises reports belongs in an algorithm register. See Algorithm registers.
- The resident knows it is AI when communicating with an assistant. See AI in citizen contact.
How werqly fits in
werqly does not replace your reporting system. It supplies the language model that runs alongside it, on a dedicated node: categorising, interpreting photos, spotting duplicates and drafting feedback, through a connection to the system you already use. Reports and photos stay in your own infrastructure. See AI for local government.
Frequently asked questions
Questions on this topic
Do authorities already use AI for public space reports?
Yes. Amsterdam categorises reports with a machine learning model in Signalen, which is available as open source and used by more Dutch municipalities. How it works is described in the Dutch Algorithm Register.
What does a language model add to an existing reporting system?
Better categorisation of vague or long reports, interpreting photos, recognising duplicate reports by content and clear feedback to the resident. The system itself stays the same.
Does such an application belong in an algorithm register?
If the model decides where a report goes or how quickly it is handled, yes: it then affects residents. Describe purpose, data, workings and who is responsible.
What privacy risks do reports carry?
Names, addresses and number plates of third parties in the text, and faces and plates in photos. Process them in your own environment, limit what is kept and delete photos once the report is closed.
Sources
- Dutch Algorithm Register — Public space reports: Signalen (City of Amsterdam, Dutch)
- VNG — Signalen (Dutch)
This article is general information, not legal advice. Rules and guidance change; check the sources listed or ask your lawyer or DPO if in doubt.