n8n as central orchestration for AI support — 10 real automations

Why an orchestration layer#

Letting Chatwoot call ERP directly and ERP webhook Chatwoot back scatters logic across systems. A few months in, nobody can untangle it.

n8n consolidates these flows into versioned, debuggable workflows.

10 real automations#

1. Ticket → CRM sync#

Chatwoot ticket → n8n webhook → create Lead in Salesforce or Lark sheet → write Lead ID back to Chatwoot note.

2. Lark / Slack alert for high priority#

Priority field triggers → n8n checks keywords (complaint, refund, urgent) → push to channel + @on-call.

3. Refund amount cross-check#

Dify Agent calls n8n webhook → n8n hits ERP → returns structured amount → Dify uses it verbatim.

4. Offline customer email follow-up#

Chatwoot conversation idle > 24h → n8n triggers → send follow-up via Postmark → reply lands back in Chatwoot.

5. Monthly CSAT report#

Cron on the 1st → n8n pulls Chatwoot conversations API → computes CSAT, deflection → generates PDF → pushes to Notion and CEO inbox.

6. KB auto-sync#

GitHub Action → n8n webhook → fetch latest docs → call Dify Datasets API → trigger eval → rollback on failure.

7. Sensitive keyword → human#

Chatwoot pre-submit hook → n8n keyword dictionary → matches escalate immediately + alert compliance.

8. Multi-storefront routing#

Ticket arrives → n8n reads customer attributes (which store) → route to corresponding inbox / agent group.

9. Customer tiering#

Nightly cron → pull Chatwoot history + CRM orders → score VIP / standard / churn → write back as Chatwoot custom attribute.

10. Human-edit feedback loop#

Agent edits an AI draft → Chatwoot webhook → n8n diffs → writes into Notion DB for the prompt engineer’s Monday review.

Where n8n sits#

Deployment notes#

• Single container is fine (4 vCPU / 8 GB)
• Use PostgreSQL, not default SQLite, for persistence
• Heavy webhook load: split webhook worker from main
• Wire critical workflows to PagerDuty / Lark for error alerts