Incoming mail and the case system: AI that registers, routes and finds the case
Short answer
Every public body receives mail, email and forms daily that must be registered, linked to a case and delivered to the right handler. That work is repetitive and error-prone. AI can classify documents by case type, extract data, find the existing case and propose a route, after which an employee confirms. It brings shorter lead times and fewer lost documents, without AI deciding anything.
Where it goes wrong
An objection arrives at the general email address, an addition to a permit application by post, a complaint through the contact form. Each document has to be recognised, registered, attached to the right case and passed on. When that fails, a deadline starts running without anyone knowing, or an applicant waits for an answer that sits with the wrong department.
What AI does at each step
| Step | What AI does | What the employee does |
|---|---|---|
| Recognise | Determines the document type: application, objection, consultation response, complaint, invoice | Assesses exceptions |
| Extract | Pulls name, address, case number, date and subject from the text | Checks when in doubt |
| Link | Finds the existing case, or proposes creating a new one | Confirms |
| Route | Proposes department and handler, with the reason | Assigns |
| Deadlines | Flags that a document starts a deadline | Schedules the deadline |
This is the kind of process to start with: high volume, a clear outcome and a mistake that is noticed early. See Where to start with AI automation.
What to watch for
- Confirm, do not pass through automatically. Start with a proposal an employee confirms; measure how often it is changed, and only then let high-confidence categories through automatically.
- Permissions follow the case system. Whoever may not see a case may not see its documents either. See Access control for AI.
- Records. A registered document is a record; its classification partly determines the retention period. See Archiving AI outputs.
- Manipulation. Incoming mail is text from outside. A letter with hidden instructions must never make an automation act; let AI only classify and propose.
- Personal data. Everything that arrives contains it. Processing belongs in your own environment.
How werqly fits in
werqly does not replace the case system. It supplies the automation alongside it, on a dedicated node: recognising, extracting, linking and routing, through a connection to the case system you already have. Documents stay in your infrastructure, the automation runs with its own identity and limited rights, and the audit log shows which document went where and why. See AI for local government.
Frequently asked questions
Questions on this topic
Can AI register incoming mail automatically?
AI can recognise the document type, extract data, find the case and propose a route. Start with confirmation by an employee; only let high-confidence categories through automatically once the figures justify it.
What happens when it makes a mistake?
A misrouted document ends up with the wrong department and is sent back; with confirmation up front that rarely happens. Measure how often employees change the proposal and use that as your quality measure.
What about deadlines that start running?
Have AI flag precisely that a document starts a deadline, such as an objection or an application, so the deadline is scheduled at once. That is one of the biggest gains of good mail handling.
Can an incoming letter abuse the AI?
Text from outside can contain hidden instructions. So let the automation only classify and propose, never carry out actions itself such as sending or deleting.
Sources
- Dutch Public Records Act 1995 (Archiefwet)
- OWASP — Top 10 for LLM Applications 2025 (prompt injection)
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.