Support triage
Use the Evaluation API with support triage playbooks to score severity, detect escalation signals, and route tickets before an agent reads the queue.
Outcome
Urgent issues escalate immediately; routine tickets stay in queue with a recommended next_action.
Prerequisites
- Deploy a Support Triage (or similar) evaluation playbook
- Live deployment +
ech_dw_…secret - Ticket or email source (Zendesk, Intercom, forwarded email, n8n webhook)
Architecture
New ticket / email
→ POST /v1/evaluate
→ status + next_action
→ ESCALATE → PagerDuty / Slack #support-urgent
→ QUALIFIED (routine) → assign tier-1 queue
→ PARTIAL → ask customer for logs / account ID
Request shapes
Ticket (input_type: "ticket")
Use when your playbook expects ticket-shaped input:
{
"input_type": "ticket",
"input": [
{
"role": "user",
"content": "Subject: Production API down\n\nWe cannot process payments since 09:00 UTC. Error 503 on all endpoints. Account: acme-prod."
}
],
"options": { "request_next_action": true }
}
Email (input_type: "email")
{
"input_type": "email",
"input": [
{
"role": "user",
"content": "From: customer@example.com\n\nBilling question: can we add 5 seats mid-cycle?"
}
],
"options": { "request_next_action": true }
}
If unsure which input_type your deployment expects, run a test in the dashboard Test evaluation panel and check the playbook docs, or use transcript with the full message body.
Branching
| status | Typical handling |
|---|---|
| ESCALATE | Immediate human + status page update |
| QUALIFIED | Standard queue with suggested priority |
| PARTIAL | Auto-reply requesting missing diagnostics |
| FAILED | Close as spam / out of scope |
next_action may include values like ESCALATE_NOW depending on your manifest.
Connect your helpdesk
| Channel | Pattern |
|---|---|
| Webhook | Helpdesk → n8n/Zapier → Evaluate → update ticket priority |
| Email forwarding | Inbound parser → HTTP POST with input_type: "email" |
| API poll | Scheduled job → fetch new tickets → evaluate batch |
See n8n integration or Zapier & Make for HTTP wiring.
Verify
- Dashboard → deployment → Test evaluation with
input_typeticket or email. - Confirm
statusandnext_actionmatch expectations for urgent vs routine samples. - Wire automation only after logs look correct in Recent evaluations.
Troubleshooting
| Issue | Fix |
|---|---|
| Wrong severity | Include subject line and account IDs in content. |
| Slow responses | Keep ticket body concise; evaluation is synchronous. |
| input_type rejected | Match playbook manifest; try transcript as fallback. |