Plan requirement
| Subscription | Suite Professional or higher, Support Professional or higher |
| Access | Admin |
Triage puts a subject, a language and a sentiment on incoming tickets. Quiet, and the best return for the least work.
The labels come with the plan; acting on them does not. Intelligent triage classifications are included from Suite and Support Professional upwards. Using them in a workflow, so a trigger or automation acts on a topic or a sentiment, needs the Copilot add-on.
What it adds
- Topic: what the ticket is about.
- Language: what the customer wrote in, from roughly 150 detected languages. Topic and sentiment are assigned for a shorter list of 30, Dutch among them.
- Sentiment: how they sound, as one of five values from Very Positive to Very Negative.
- Entities: specific values pulled out, such as an order number.
They arrive as fields, so everything that reads a field can use them: views, triggers, routing, reporting. Zendesk adds matching tags too, which is often the simpler hook. Each classification also carries a confidence field saying how sure it is, and that is the one worth reading before you automate on it.
Four labels on incoming requests, and none of them changes the ticket. That is the whole feature, and the reason it is safe to switch on before you have decided what to do with it.
Why it is the best first step
It changes nothing a customer sees, so nothing can go visibly wrong. It needs no content preparation. And it produces, within a week, an accurate answer to the question most support teams cannot answer: what do people contact us about?
What it enables
- Routing by subject, instead of by whoever picks it up.
- Views that mean something, such as billing questions across all channels.
- Spotting the angry ones early, before they escalate.
- Reporting by subject, which is what tells you where to write articles.
Labels, not decisions
Triage adds information; it does not act on it. Nothing changes in how work flows until you build a view, a trigger or a routing rule that uses the labels.
That is a feature: you can run it for a month, see whether the labels are accurate, and only then act on them.
It is not perfect
Short tickets, mixed subjects and unusual phrasing get labelled imperfectly. Agents can correct it, and the corrections are worth encouraging.
When a ticket gets no labels
Not every ticket is classified, and five documented cases explain most of the gaps.
- It predates the switch. Triage does not go back over old tickets; only ones created after you turned it on.
- Your account does not qualify for intent detection. That depends on industry and model fit. Language and sentiment still arrive; the topic does not.
- The channel or route is not covered. Follow-up tickets to closed tickets, and tickets created through the Ticket Import API, get nothing.
- It opened with an internal note. Only tickets created with a public comment are classified.
- Ignore agent-initiated tickets is on. Then agent-created tickets get no labels, and neither do suspended tickets you recover.
A trigger or an integration can also overwrite a prediction after the fact. The events view on the ticket shows whether a label arrived and was later removed.
Comments
0 comments
Article is closed for comments.