AI-powered ops workflow with n8n and Claude
Built an automation pipeline that routes incoming support requests, drafts responses using Claude, and updates the ticketing system — cutting average response time from 4 hours to under 15 minutes.
n8nClaude APIPythonAutomationAI
The situation
A service business was handling support requests manually: someone read the email, decided what type it was, drafted a reply, updated the CRM, and assigned it to the right person. This happened 30-50 times a day. It was the kind of work that takes a person 10 minutes each time but could be automated in an afternoon.
What I delivered
Intake pipeline (n8n):
- Webhook listener on incoming email/form submissions
- Classifier node: routes by request type (billing, technical, onboarding, general)
- Priority scoring based on keywords and sender history
AI response drafting (Claude API):
- System prompt with company voice and policy guardrails
- Draft generation per request type — not a generic reply, a context-aware one
- Confidence threshold: low-confidence drafts flagged for human review before sending
CRM integration:
- Automatic ticket creation with classification, priority, and draft attached
- Human-in-the-loop approval step for sends over a defined threshold
- Logging of all AI-assisted actions for audit purposes
Outcome
- Average response time: 4 hours → under 15 minutes
- 80% of routine requests fully automated (human reviews the other 20%)
- Staff freed from triage work — they focus on the requests that actually need a person
- Full audit trail of what the AI sent and when
Stack: n8n self-hosted, Claude Haiku for classification and drafting, Python for the CRM connector. Client details withheld.