TOPdesk × AI: connecting AI agents to your ITSM
TOPdesk is the most-used ITSM platform in the Netherlands, with over 4,500 customers in the Benelux. This guide shows how to connect AI agents to TOPdesk to automatically classify tickets, speed up knowledge base lookups, and execute repetitive actions — without disrupting your existing TOPdesk setup.
What you gain with AI agents on TOPdesk
An AI agent on top of TOPdesk adds three capabilities without replacing TOPdesk itself:
- Automatic ticket classification — inbound tickets get the right category, urgency, impact, and assignee group within seconds. Saves 30-90 seconds per ticket on service desk triage.
- AI-assisted draft reply — based on ticket content and relevant knowledge base articles, the agent drafts a reply that the handler approves or edits.
- Automatic knowledge capture — resolved tickets with valuable outcomes are turned into knowledge base article drafts (for editorial review), so your knowledge base grows with you.
Technical architecture
The integration runs via the TOPdesk REST API plus TOPdesk Event Actions. No on-premise component is needed; TOPdesk SaaS is fully supported.
| Component | Role |
|---|---|
| TOPdesk Event Action | Triggers on new ticket, status change, new note |
| Webhook endpoint | Receives events from TOPdesk and forwards to the AI agent |
| AI agent runner | Classifies, drafts replies, retrieves knowledge |
| TOPdesk REST API (PATCH/POST) | Updates the ticket with the AI suggestion or takes action |
| Audit log | Records every AI decision for compliance |
Step 1: Create an operator account
In TOPdesk, create a dedicated operator account for the AI integration:
- Go to Settings → Functional settings → Login settings → Application passwords
- Create a new application password named
ai-servicechanger - Assign the minimum required role: read tickets, add replies, mutate category/handler. No user management or settings permissions.
- Store the password securely (you see it only once)
Tip: don't reuse an existing operator account. A dedicated account keeps audit trails clean and you can revoke it anytime without impacting your staff.
Step 2: Register the webhook endpoint
In ITSM Autopilot: configure a new TOPdesk environment and paste the application password. The platform returns a webhook URL:
https://api.itsmautopilot.com/webhooks/topdesk/<your-tenant-id>Step 3: Configure Event Actions
In TOPdesk: Settings → Event management → Event Actions → new.
Start with at least three events:
| Event | Trigger | Purpose |
|---|---|---|
incident.created | New incident created | Classification + knowledge base check |
incident.updated | Status change or note added | Re-evaluation if the user added info |
incident.closed | Ticket closed | Trigger for knowledge article draft |
application/json.
Step 4: Shadow mode (required!)
Never put AI agents live on production tickets directly. ITSM Autopilot starts in shadow mode by default: the agent reads along, makes suggestions, but doesn't mutate tickets. You see in the dashboard what the agent would have done.
Run at least 2 weeks of shadow before enabling autonomous mode per agent rule. This reveals:
- Which ticket categories the AI classifies well (usually >95% after a week)
- Where the knowledge base has gaps (AI can't find a solution = knowledge base incomplete)
- Which edge cases will need human review forever
Step 5: Gradual rollout
Once shadow results convince you: enable autonomous mode per category, not globally. Order from low to high risk:
- Classification (mutates category/urgency, not the user experience)
- Knowledge article suggestion (shown to handler, not to user)
- Draft reply (handler approves)
- Autonomous reply (only for categories with 99%+ confidence)
- Autonomous tool actions (account unlock, group membership) — only after months of proven track record
Pitfalls
| Problem | Cause | Fix |
|---|---|---|
| AI classifies everything as "Other" | Category descriptions missing or insufficient in TOPdesk | Add 1-2 examples per category in the category description |
| Knowledge articles aren't used | Articles in draft status, not published | Filter on published + fill out metadata (tags, category) |
| Duplicate actions on status change | Multiple Event Actions triggering on the same update | Add a filter in the TOPdesk Event Action: only on specific status transitions |
| Webhooks fail intermittently | TOPdesk has no built-in retry | ITSM Autopilot retries with exponential backoff; check the event log |
Performance metrics
Measure from day 1, also in shadow:
- Classification accuracy (AI-suggested category vs final category)
- MTTR (mean time to resolve) per ticket category, before vs after AI
- First Contact Resolution rate for categories where AI drafted a reply
- Service desk handler-time per ticket (via time entries in TOPdesk)
Frequently asked questions
Does this work with on-premise TOPdesk? Yes. The integration is API-based, so on-premise works as well as SaaS. The webhook endpoint must be reachable from TOPdesk (public URL or reverse proxy through your firewall).
What happens to ticket content? Does it go to external LLMs? No, not by default. ITSM Autopilot runs on EU-hosted LLMs (Azure OpenAI EU or Mistral); ticket content doesn't leave the EU. For customers with stricter requirements, a self-hosted LLM is also possible.
What about personal data (GDPR)? Before the prompt, PII is automatically masked (names, email addresses, customer numbers) unless the use case explicitly requires it. See also our article on PII masking in the service desk.
Can we start without commitment? Yes, 30-day free trial without credit card. After 30 days you pick a plan or stop — data is deleted within 7 days.
What now?
Setting up a working TOPdesk integration typically takes 1-2 days (account + events + shadow configuration), then 2 weeks of shadow before you're comfortable. Schedule a demo or start the trial directly.