n8n in production - what we learned after 30 implementations

n8n works well as a place to prototype, but production calls for a different kind of discipline. After 30 implementations we know what separates a workflow that simply runs from one that calls you at three in the morning.

Five standing rules

1. Self-hosted from day one. n8n Cloud is fine for an MVP, but a production system with an SLA has to sit on your own infrastructure - control over upgrades, auditing, GDPR.
2. Every workflow has an error workflow. We never leave an execution without a catch-all - a failure has to reach Slack or the ticketing system.
3. Retries with exponential backoff. Three attempts, 5s/30s/120s. 90% of failures are transient timeouts.
4. Webhook authentication, always. Even when the endpoint is “internal”. An HMAC signature or a JWT.
5. Version control through git. JSON exports into the repository, plus CI that imports them into staging automatically.