Route support messages with GPT-4.1-mini, Slack, email and human review

Description

This workflow classifies incoming support messages from email or Slack, drafts AI-generated replies for routine queries, and automatically pauses to notify a human agent for sensitive topics such as complaints, emotional distress, billing disputes, legal concerns, or escalation requests.

Who's it for
• Customer support teams handling 50+ messages per day
• SaaS companies needing compliant, empathetic AI responses
• E-commerce brands managing returns, refunds, and complaints
• Any team that wants AI speed with human oversight on sensitive issues

How it works / What it does
Captures incoming messages from Email (webhook) or Slack
Normalizes message context (sender, channel, body, urgency signals)
AI classifies the message into a category and sensitivity tier
Routine queries → AI drafts reply → sent automatically
Sensitive queries → AI drafts reply → paused for human review
Human approves, edits, or rejects the draft via webhook resume
Final reply sent to customer via email or Slack
All interactions logged to Google Sheet with category, tier, and resolution

How to set up
Import this workflow
Configure credentials (Gmail/IMAP, Slack, OpenAI/Anthropic, Google Sheets)
Update your support tone, escalation keywords, and team Slack channel
Activate workflow

Requirements
• Gmail or IMAP webhook / Slack Events API
• OpenAI / Anthropic / Grok API
• Slack Bot Token (for human notifications)
• Google Sheets for audit logging
• HTTP endpoint for human approval webhook

How to customize the workflow
• Edit sensitivity tiers in the Python classification node
• Change escalation keywords list
• Update Slack notification channel and message template
• Modify AI tone and persona in the AI draft node
• Adjust wait timeout for human review window