Auto-tag and route support tickets

n8n plus ChatGPT sends each ticket to the right queue.

Peter OseiAdvancedSaves ≈5h/weekn8nChatGPT
366 views
Free community read · 5 leftUnlock all →

Triage was a full-time chore. Now tickets land in the right place with a suggested priority.

The workflow

  1. Trigger n8n on a new ticket
  2. Send the body to ChatGPT for a category and priority
  3. Route to the matching queue
  4. Notify the owner

Result: Tickets triaged the second they arrive.

The prompt

Prompt
Classify this support ticket. Output category, priority (P1 to P4), and a one line reason. Ticket:
[PASTE]

A note on what follows. The workflow, the prompt and the result above come from the member. The notes below (why the prompt works, what to watch for, how to adapt it) were written with AI help.

Why this prompt works

The one line reason is doing more work than the category. Early on I only asked for category and priority, got a clean P2, and had no way to tell whether it understood the ticket or matched a keyword. Now the reason is right there and I can dismiss a bad call in a second. Forcing P1 to P4 rather than "high, medium, low" also helped, because our team already argues in those terms and vague labels drifted week to week. What I got wrong at first was pasting only the customer's message. Adding the account tier and how long the ticket has been open changed the priorities completely, which in hindsight is obvious.

What to watch for

  • Angry tone inflates priority. A polite ticket about data loss can land below a furious one about a typo.
  • It cannot see history. A fifth ticket from the same customer about the same thing looks like a first ticket.
  • Multi-issue tickets get one category, and the second issue quietly disappears. We now split those by hand before running it.
  • New product areas get forced into old categories until you tell it the new one exists.

How to adapt it

  • Pin the taxonomy, otherwise it invents categories. Add Use only these categories: billing, access, data, integrations, bug, how-to.
  • For tiered support, add The account is on the enterprise plan with a 4 hour response SLA. and watch the priorities move.
  • If you route automatically, add If confidence is below high, output needs-human instead of a category. We only automated once that was in place.

What good output looks like

The reason line should quote something from the ticket, not restate the category. "P1 because the customer reports production data missing since this morning" is useful. "P1 because this is urgent" means it pattern matched on a word. I also expect a handful of needs-human every day. When a week goes by with none, the confidence rule has stopped biting and I recheck a sample by hand.

Want to share your own workflows?

AI Academy members publish workflows, vote, and get featured in Techpresso.

Become a member

Discussion

Loading…