AI Ticket Classification
Helpdesk uses AI to automatically classify incoming tickets across multiple dimensions. Classification runs on every new ticket and helps agents prioritize, route, and respond faster.
Classification fields
| Field | Description |
|---|---|
| Spam score | Likelihood the ticket is spam or promotional (0-100) |
| Automation score | Likelihood the ticket is an auto-reply or out-of-office (0-100) |
| Sentiment | Detected emotion: positive, neutral, negative, or frustrated |
| Intent | What the customer is trying to accomplish (e.g. billing inquiry, feature request) |
| Language | Detected language of the ticket |
| Suggested priority | AI-recommended priority level based on content |
| Detected skills | Skills extracted from the ticket for routing |
| Confidence | Per-field confidence scores indicating how certain the AI is |
How classification works
- A new ticket arrives via email, form, or API
- The AI model analyzes the subject and body
- Classification results are attached to the ticket
- Agents see the results on the ticket detail page
- Automations and triggers can use classification fields as conditions
Classification runs asynchronously. If the AI service is temporarily unavailable, tickets are queued for retry via the backfill queue.
Using classification in automations
AI classification fields can be used as conditions in triggers and automations. For example:
- Auto-close tickets with a spam score above 90
- Escalate tickets with negative sentiment to a senior group
- Route tickets by detected intent to the appropriate team
- Auto-tag tickets based on detected language
AI-powered authoring
In addition to classification, Helpdesk offers AI-assisted email composition:
- Compose — Generate a draft reply from a short intent description. Optionally include ticket context and knowledge base articles for more relevant responses.
- Rewrite — Transform existing text with options to rewrite, expand, shorten, fix grammar, or change tone.
Both features are rate-limited to 20 requests per minute per agent.
Related
- Intent Categories — customize intent classification
- Agent Corrections — improve AI accuracy with feedback
- Rules & Triggers — use AI fields in automation conditions