What this automation does
When a support ticket is transferred between agents — shift changes, escalations, or follow-ups — the new agent has to read through the entire conversation history. For complex issues, that can be 20+ messages.
This automation generates a concise summary: what the customer's issue is, what solutions have been tried, what worked or didn't, and what the customer expects next. The summary appears as an internal note at the top of the ticket.
Tools you need
- Zendesk, Freshdesk, or Intercom: Your support platform with API access
- OpenAI API: For summarizing conversation threads ($0.02-0.05 per summary)
- Zapier or Make: Triggers on ticket reassignment and posts the summary as an internal note
How to set it up
Step 1: Set up a Zapier trigger that fires when a ticket is reassigned or when a ticket has been idle for a set period. Pull the full conversation history via the helpdesk API.
Step 2: Send the conversation to OpenAI with a prompt like: 'Summarize this support conversation in 3-5 bullet points. Include: (1) the customer's core issue, (2) what has been tried, (3) current status, (4) what the customer is expecting next.'
Step 3: Post the summary as an internal note on the ticket, visible only to agents. Tag it clearly so agents know it's an AI-generated summary.
Step 4: Optional — also generate a suggested next action for the agent. This helps new agents pick up tickets faster.
Cost breakdown
| Item | Cost | Notes |
|---|---|---|
| Zapier or Make | $10-$20/mo | Low volume of triggers |
| OpenAI API | $5-$15/mo | Summaries use more tokens but fewer calls |
| Setup time | 30-45 min | One-time |
| Total monthly | $15-$35/mo | Saves 3+ min per ticket handoff |
Frequently asked questions
Yes. GPT-4 handles up to 128K tokens of context — that's roughly 200+ messages. For extremely long threads, truncate to the most recent 50 messages, which covers the vast majority of cases.
The summary is posted as an internal note, so agents can add their own notes below it. If the summary misses something important, the agent can append a correction.