Retry-safe lead intake in n8n: duplicate suppression, human review and CRM 503 recovery

I built and tested a small n8n capability proof at Row & Rule around a failure mode I keep seeing in intake workflows: a lead is accepted, a downstream write becomes uncertain, and a blind retry creates a second CRM record or task.

Important context: this uses synthetic events and no credentials or client data. It is an executed demonstration, not a claim about a client production deployment.

What it does

  • normalises each incoming event and creates a stable identity/idempotency key;
  • validates required fields before any downstream write;
  • suppresses a replay before CRM or task creation;
  • routes an ambiguous region to explicit human review;
  • assigns priority and an SLA deadline;
  • deliberately simulates a CRM 503, retries with the same key and recovers without downstream duplicates;
  • upserts the master sheet and project board, then writes an audit-ready outcome.

Test run on n8n 1.113.3

  • 4 synthetic events;
  • 3 delivered;
  • 1 replay suppressed;
  • 1 human review approved;
  • 1 CRM recovery on attempt 2;
  • 0 duplicate CRM records;
  • 0 duplicate project tasks.

Limits

This run proves the visible paths in one local execution. It does not prove distributed-concurrency safety. In a production version, I would persist claim state outside execution memory, use an atomic claim or processing lease, and reconcile with the destination before retrying any write whose outcome is uncertain.

77-second walkthrough: https://rowandrule.com/assets/n8n-lead-intake-proof-en.mp4

Importable workflow JSON: https://rowandrule.com/assets/row-and-rule-n8n-lead-intake-sla.json

For HTTP/API nodes, when the destination may have accepted a write but n8n receives a timeout or an uncertain response, which pattern has been most reliable for you in production: a stable idempotency key plus destination reconciliation, an external claim table with a lease, or something else?