n8n workflow automation

Most automations work fine until 2am on a Tuesday, when an API hiccups and your lead data silently vanishes. This repo shows the three importable n8n workflows we use to prevent that — every step retries, every failure lands somewhere visible, and nothing is ever dropped quietly.

Built by Vroom Analytics as public proof for our AI automation & orchestration service.

---

What's inside

Three realistic n8n workflows, plus the Python that proves they hold up:

File	What it does	The 2am story
workflows/lead-intake-pipeline.json	Webhook → qualify → CRM write → confirmation	CRM write retries 3×, then dead-letters the lead and alerts a human
workflows/nightly-crm-sync.json	Midnight schedule → sync records in batches	One bad record can't kill the sync; it parks the record and keeps going
workflows/dead-letter-reprocessor.json	Every 15 min → retry parked items	Reprocesses failures; poison messages get quarantined and paged, never retried forever
src/n8n_reliability/validator.py	The reliability checklist as code	4 checks every workflow must pass (below)
src/n8n_reliability/retry.py	Exponential-backoff retry policy	The Python mirror of n8n's retryOnFail
src/n8n_reliability/dlq.py	File-backed dead-letter queue	Park → requeue → done / quarantined, persisted as JSON

Every workflow passes the same four checks, enforced by validator.py:

1. Valid n8n shape — nodes with id/name/type/parameters/position, connections that reference real nodes.
2. Error handling — an errorWorkflow, an Error Trigger node, or onError routing. The n8n default (stop the workflow, lose the data) is not acceptable.
3. Retry config — every node that touches an external system sets retryOnFail with explicit maxTries >= 2 and waitBetweenTries. Explicit, not defaulted: retry settings you can't see are settings you can't audit.
4. A failure path — failures route to something wired: a dead-letter, alert, or quarantine node. An unwired node with "dead letter" in its name is decoration, not a failure path.

5-minute setup

All steps below were run verbatim on a clean machine — they work as written.

1. Get the code and enter the folder:

   cd n8n-ai-workflow-automation

2. Create a virtual environment and install the one dependency:

   python3 -m venv .venv
   source .venv/bin/activate
   pip install -r requirements.txt

3. Run the reliability gate over all three workflows:

   python - <<'EOF'
   import sys
   sys.path.insert(0, "src")
   from n8n_reliability import validate_workflow_file
   for f in ["lead-intake-pipeline", "nightly-crm-sync", "dead-letter-reprocessor"]:
       r = validate_workflow_file(f"workflows/{f}.json")
       print(f"{r.name}: {'RELIABLE' if r.is_reliable else 'FAILED'}")
   EOF

   Expected: all three print RELIABLE.

4. Import one into n8n: in n8n, go to Workflows → ⋯ → Import from File and pick e.g. workflows/lead-intake-pipeline.json. Replace the https://crm.example.com URLs and the #leads / #ops Slack channels with your own, add your credentials to the HTTP/Slack/Email nodes, and activate.

Running the tests

pip install -r requirements.txt && pytest -q

Expected: 88 passed. The suite covers the failure modes this repo exists to prevent:

• A workflow with no error handling does not validate
• A workflow with no retry config on a network node does not validate
• A workflow with no failure path does not validate (including the sneaky case: a dead-letter node nobody routes to)
• Poison messages are quarantined after the requeue limit — never retried forever, never silently re-queued
• Retry backoff follows the exponential schedule; permanent errors raise immediately instead of burning retries
• Dead-lettered items survive restarts (the queue is file-backed)

Plus: schema validation of every node/connection, FIFO reprocessing order, and audit records for everything the queue touches. No network calls in tests — everything is fixture-backed.

How this maps to the offer

This is the engineering behind our AI automation & orchestration service (S01): the Basic package's "error handling, failure alerts," the Standard package's "monitoring view + alert routing," and the Premium package's "30-day monitoring." Buyers don't pay for the workflow — they pay for never having to think about it at 2am. These workflows and the validator are the proof we actually build that way.

The Vroom page: AI automation & orchestration — vroomanalytics.com/automation

Monthly peace of mind

Setup is just day one. The $99/mo care plan keeps this running — monitoring, fixes, and monthly optimization, so you never think about it again.

License

MIT — see LICENSE. Steal the dead-letter pattern; it's the cheapest insurance in automation.