Overview and setup
What the three agents do, the order to switch them on, and what you need first.
What it is. Three separate features, each of which you enable independently. You can use one and ignore the others.
| Agent | When it runs | What it does |
|---|---|---|
| Dispatcher Agent | A ticket is created | Chooses the team, then the person, and assigns the ticket. Optionally leaves a comment explaining why. |
| Smart Escalation | A customer adds a comment | Scores how negative the comment is. Past your threshold it can reassign, add a watcher, comment internally or add a label. |
| Incident Detection | A ticket is created, plus a recurring scan | Groups tickets about the same problem, promotes one of them to major incident, and links the rest to it. No new issue is created. |
Who sets it up. A Jira administrator for the global configuration. Teams that own their own service desk can configure their desk themselves, if it is not already covered globally.
What you need first
- Jira Service Management for the service desk features. The app is installed into Jira.
- Jira administrator rights.
- For team-based routing only: an Atlassian account with organisation access and an API token, so the app can read your Atlassian Teams. Everything else works without it.
- Nothing to buy for the AI. It runs on Atlassian's Forge LLM.
Switch them on in this order
This order exists so you can see the app's judgement before it changes anything important.
- 1Install the app, open Jira → Apps → AI Triage → Connection, and enter the Atlassian Teams credentials. Do this once; every project uses them.
- 2Open one service desk project and go to Project settings → AI Triage → Dispatcher Agent. Switch Enable Dispatcher Agent on.
- 3Select the teams and write a routing prompt for each one. Leave Use Team Routing Prompts on if you want it to pick a person as well as a team; turn it off and it assigns the team only.
- 4Watch it for a few days. Read the AI Triage panel on routed tickets and the global Audit Log — the app never writes a comment on a ticket, so those two are your evidence. When it gets one wrong, fix that team's prompt. The prompt is where accuracy comes from.
- 5Move to the Smart Escalation tab. Enable it, leave Sentiment Threshold at Negative or worse (recommended), and select no actions. The scores appear in the issue panel and the audit log without anything changing on the ticket.
- 6When the scores match what your team would have judged, add the actions you want — Flag with label first, then reassignment and the watcher once you trust it. See the Smart Escalation page.
- 7Enable Incident Detection last, with a cluster threshold high enough that it only fires on a real wave.
- 8Repeat for the next project.
Two screens, and which one does what
This is the thing that trips people up, so get it straight before you start. The agents are configured per project. The global page only holds the shared Atlassian Teams credentials, the statistics and the audit log.
| Screen | Tabs | What you do there |
|---|---|---|
| Jira → Apps → AI Triage (global) | Statistics · Connection · Audit Log | Enter the Atlassian Teams API credentials once, for the whole site. Review statistics and the audit log. |
| Project settings → AI Triage (per project) | Statistics · Dispatcher Agent · Smart Escalation · Incident Detection | Enable and configure each agent for that project's service desk. This is where you spend your time. |
| The AI Triage panel on an issue | — | See what the agents decided and why. Anyone who can see the issue. |
| Personal settings → My Skills | — | Each agent describes their own specialities, for routing. |
AI Triage
Configure the Atlassian Teams API credentials used by the Dispatcher Agent to access organization teams and members. This is a global setting shared across all projects.