Alec AsgariThe first ten n8n workflows are fun. Around workflow thirty, things change: nobody remembers which...
The first ten n8n workflows are fun. Around workflow thirty, things change: nobody remembers which "Webhook 3" handles which form, the same Slack-notification block has been copy-pasted into twelve places, and changing one API key means opening half the instance. None of this is n8n's fault. It's what happens to any automation setup that grows without a few conventions.
These are the house rules I set up for small teams running self-hosted n8n. None of them need extra tools.
A consistent pattern makes the workflow list searchable:
[AREA] Trigger -> Outcome
For example:
[SALES] Website form -> CRM lead + Telegram alert
[FIN] Daily 08:00 -> Unpaid invoice reminders
[OPS] Supplier email -> Parse PDF -> Sheet
[LIB] Send Telegram alert
The [LIB] prefix marks reusable sub-workflows (more on those below). Tags work too, but a name prefix is visible everywhere, including execution logs.
HTTP Request1 and IF2 mean nothing three months later. Rename nodes to say what they do: Get open invoices, Has email?, Format reminder text. When an execution fails, the error points straight at a readable step.
If the same group of nodes appears in more than two workflows, it should probably be its own workflow, called with the Execute Workflow node. Typical candidates:
Give the sub-workflow an Execute Workflow Trigger with a clear input shape, and document it in a sticky note:
{
"channel": "telegram",
"level": "error",
"title": "Invoice sync failed",
"message": "Supplier API returned 503"
}
Now when you switch alert channels or change message formatting, you change one workflow instead of twelve.
Avoid five slightly different credentials for the same Google account. Name credentials by system and purpose, like Google Sheets - finance (service account), and reuse them. When a token rotates, there's one place to update.
Hard-coded IDs scattered through nodes (sheet IDs, chat IDs, folder IDs) are a common source of "it worked in testing". Options that work well on a small setup:
$env.MY_VAR where allowed[LIB] Get config sub-workflow that returns a JSON objectPick one and stick with it.
Set an Error Workflow in each production workflow's settings and point it to one shared handler that sends the workflow name, the failing node and the execution URL to your alert sub-workflow. A silent failure is the most expensive kind.
A short sticky note at the start of each workflow answering three questions saves a lot of time:
n8n workflows are JSON. A nightly job that exports them via the n8n API or CLI and commits them to a private repo gives you history, diffs and a backup for free. It also makes reviewing changes possible when more than one person edits automations.
Even a second, small n8n instance (or at least a [TEST] prefix and test credentials) keeps experiments from sending real emails to real customers.
None of these rules is complicated, and you don't need all of them on day one. Naming, error workflows and one or two shared sub-workflows already make a big difference. I've written more about moving teams onto self-hosted n8n in this Zapier-to-n8n migration guide, and I set up and maintain these systems for small businesses through my n8n automation service.