Try it with your own rules
Start from the preset below, adapt the questions, and inspect the typed answers. Run it live in the playground — it is the same request shape your application will send.
What Jev returns
Illustrative output for the preset below — run it live to get real values for your input.
Since yesterday every CSV export stops halfway. I retried three times — our customer review is this afternoon. Can someone help before then?
Which team should handle this support ticket?
Does this message describe a problem that needs attention today?
P(yes) = 0.94 — explicit deadline this afternoon
How frustrated does the customer sound?
From one example to a reusable workflow
Keep output fields aligned with queue operations
Ask for the queue as a choice over your real team names, severity as a score, and escalation as a probability — fields that map straight onto ticket fields.
Separate urgency from topic
A billing question can be urgent and a bug can wait — ask both, don't let the topic decision smuggle in priority.
Measure queue health, not confidence
Time-to-first-response, escalation precision and reopen rate tell you if the triage helps — a confident misroute is still a misroute.
Keep the criteria separate
| Check | What it measures | How to use it |
|---|---|---|
| Queue · Choice | Which team owns this ticket? | Option definitions are the routing policy — edit them as teams and products change. |
| Escalate · Yes/No | Does this need a human today? | The noul threshold sets your false-escalation cost: 0.8 keeps quiet days quiet, 0.5 catches more deadlines. |
| Frustration · Score | How heated is the customer? | Route high-frustration tickets to your best agents regardless of topic. |
The problem
Support teams combine keyword rules, manual review and a general model — hard to tune, easy to over-escalate. Typed decisions replace three fuzzy signals with fields your ticketing system already understands.
The production loop
Log queue, escalation probability and downstream resolution. A good triage classifier improves the queue measurably — it is not one that merely looks confident on the dashboard.
Cheap enough to run on every ticket
One request with three questions bills a few hundred input tokens — decision models are priced so every ticket gets judged, not just the ones a human sampled.
FAQ
Can triage handle multilingual tickets?
The same request shape works on any language your model supports — ask the questions in English and let Jev read the ticket.
How is this different from the intent preset?
Intent routing picks what a chatbot should answer; triage picks which human queue owns the ticket and how hot it is.
What if the ticket fits two queues?
The probabilities map shows the split — 0.6/0.4 tickets are exactly the ones worth a second look or a broader queue.