What this automation does
Every incoming support ticket gets read by AI, which classifies it into categories (billing, technical, shipping, account, etc.) and assigns an urgency level. The ticket is then automatically routed to the correct team or agent queue.
This eliminates the bottleneck of manual triage, which in most companies is handled by a single person scanning every ticket. With AI routing, tickets reach the right person within seconds instead of minutes or hours.
Tools you need
- Zendesk, Freshdesk, or Help Scout: Your support ticketing platform
- OpenAI API or Claude API: For ticket classification ($0.005-0.02 per ticket)
- Make (Integromat): Connects your ticketing system to the AI model and routes back
How to set it up
Step 1: Define your ticket categories. List every team or queue tickets should go to — billing, technical support, shipping, account management, partnerships, spam. Write 2-3 example tickets for each category.
Step 2: Create a Make scenario triggered by new tickets in your helpdesk. Pass the ticket subject and body to OpenAI with a prompt that includes your categories and examples. Ask the model to return a JSON object with category, urgency (low/medium/high/critical), and confidence score.
Step 3: Add a router module in Make. Based on the AI's classification, update the ticket's tags, priority level, and assignee group in your helpdesk. For critical urgency, also send a Slack notification to the team lead.
Step 4: Run in shadow mode for one week — let the AI classify but don't auto-route. Compare its classifications against your human triage. Adjust the prompt and examples until accuracy hits 90%+.
Cost breakdown
| Item | Cost | Notes |
|---|---|---|
| Make | $10-$30/mo | Based on number of operations |
| OpenAI API | $5-$20/mo | ~$0.01 per ticket at 500-2000 tickets/mo |
| Setup time | 45-90 min | One-time, plus a week of tuning |
| Total monthly | $15-$50/mo | Replaces 1-2 hours/day of manual triage |
Frequently asked questions
With well-defined categories and 2-3 examples per category, AI routing achieves 90-95% accuracy. The remaining 5-10% are edge cases that get routed to a general queue for human triage.
Yes. GPT-4 and Claude both handle multilingual classification well. Add a language detection step and route non-English tickets to the appropriate team or auto-translate before classification.