Smart Escalation
Notice when a customer is losing patience, and act before the relationship breaks.
What it does. Every time a customer comments, the agent judges how negative the comment is. If it crosses your threshold, it performs the actions you allowed.
Why. The clearest sign that a ticket is going wrong is in the customer's words, not in the SLA clock. By the time the SLA breaches, the damage is done.
Project settings → AI Triage → Smart EscalationIllustration
AI Triage
StatisticsDispatcher AgentSmart EscalationIncident Detection
Enable Smart Escalation
Automatically analyze customer comments for negative sentiment and trigger escalation actions.
▾ General SettingsCollapse
Sentiment Threshold
Negative or worse (recommended)
Escalation actions trigger when sentiment meets or exceeds this level.
Escalation Label
escalation-risk
Label added to issues when the "Flag with label" action is enabled.
Custom AI Instructions
Additional instructions for sentiment analysis...
Optional instructions appended to the sentiment analysis prompt.
▾ Escalation ActionsCollapse
Select which actions to perform when negative sentiment is detected.
Reassign to escalation user
Add escalation user as watcher
Flag with label
▾ Escalation UsersCollapse
Assign an escalation user for this service desk. This user will be used for reassign and watcher actions.
Priya Raman — Service Desk Lead
The threshold
Comments are classified as positive, neutral, negative or critical. Sentiment Threshold is the level at which actions fire, and the dropdown names the levels in plain language.
| Threshold | Effect | Use when |
|---|---|---|
| Negative or worse (recommended) | Acts early. More escalations, including some that did not need it. | Most desks. This is the default and the one we recommend. |
| Critical only | Acts only on the clearest cases. | High volumes, or while you are still calibrating and do not want noise. |
The actions
Three, and they work from a global configuration or a per-project one. Tick as many as you want, or none.
| Action | What it does | Needs |
|---|---|---|
| Flag with label | Adds the label you configure, so escalated tickets are one JQL query away. | The Escalation Label field. Nothing else. |
| Reassign to escalation user | Hands the ticket to the escalation owner for that desk. | An escalation user for that desk. Visible to the customer — see the warning below. |
| Add escalation user as watcher | Keeps the current assignee and adds a senior person as a watcher. | An escalation user for that desk. The gentlest of the three. |
The app does not comment on issues
Nothing it does writes to an issue. The score, the reasoning and the key phrases appear in the AI Triage panel on the issue itself and in the audit log, where they are attached to the analysis rather than mixed into the conversation the requester is reading.
The other settings
- Escalation Label
- The label used by the Flag with label action. It ships as
escalation-risk; keep it or pick something else you can query. - Custom AI Instructions
- Your own rules, added to the analysis. This is where domain-specific triggers go, for example: “treat contract-renewal language and legal threats as critical regardless of tone”.
- Escalation Users
- One person per service desk, used by the reassign and watcher actions. It is a user picker — search by name and choose. Configure this before ticking either of those two actions; without a user for that desk, both are skipped.
- Issue Types (optional filter)
- Analyse only some issue types. Leave it empty to analyse all of them.
- Only process tickets matching this JQL
- A JQL filter checked before the comment is analysed. Anything that does not match is ignored entirely — no model call, no score, no audit entry. Useful for excluding a noisy request type, or for piloting on part of a desk.
Reassignment is visible to the customer
It changes who they are talking to, mid-conversation. During calibration use the label and the watcher, which the customer never sees. Enable reassignment when the scores have proven themselves.
Everything on this page is per project
The three sections — General Settings, Escalation Actions and Escalation Users — are collapsible panels on the Smart Escalation tab of that project's settings. Two service desks can have completely different escalation policies, and neither needs a site-wide change.