All use cases
Customer Support Triage
Auto-classify and route support tickets. Cut response time from 4 hours to 23 minutes.
The problem
Support team drowning in tickets. 60% are repeat questions with known answers. Average first-response time sitting at 4 hours. Two senior agents spending half their day just routing tickets to the right queue.
The result
First-response time dropped from 4 hours to 23 minutes. 40% of P3 tickets now resolved without human intervention. Senior agents reclaimed ~15 hours/week.
The workflow
How the system actually runs.
- 01 Incoming ticket hits Intercom
- 02 Zapier sends ticket text to Claude API
- 03 Claude classifies urgency (P1/P2/P3) and category
- 04 Auto-routes to correct team based on classification
- 05 P3 tickets get auto-drafted response for human review
Tools used
The stack behind it.
Claude APIZapierIntercom
What worked
Why this setup held up.
- Classification accuracy hit 94% after two rounds of prompt tuning
- Auto-drafts for common questions saved the most time by far
- Adding ticket history as context improved routing accuracy significantly
What did not
The friction to watch.
- First attempt at full auto-reply caused 3 incorrect responses in one day — reverted immediately
- Multi-language tickets had lower accuracy until we added language detection step
- Zapier rate limits caused missed tickets during peak hours — had to add retry logic
Verdict
The short version.
Worth it. But start with triage only — don't auto-reply until you trust the outputs. Human-in-the-loop for anything above P3.