Search "n8n examples" and you get the same thing every time: a gallery of screenshots, a list of 15 use cases described in one sentence each, and no idea what actually happens between the trigger and the output. It looks impressive. It teaches you nothing about whether the workflow survives contact with real data.
This post is different. Below are 12 n8n examples we have actually built and run for clients, grouped by the job they do. For each one you get the node map (the sequence of nodes, not a screenshot), the place it tends to break, and a plain verdict on whether it is worth building at all. That last part is what every other list of n8n examples leaves out.
If you want the strategy layer behind sequencing builds like these, our workflow automation services page covers how we score and prioritize them. This is the field guide to the workflows themselves.
How to read these n8n examples
Every workflow below follows the same format so you can scan it fast:
- Node map: the ordered sequence of nodes, written as a list
- Breaks when: the specific failure mode that shows up in production
- Verdict: build it, buy it, or skip it
A quick note on one distinction that matters. Some of these n8n examples are pure rule-based automation - trigger, transform, send. Others need an AI node in the middle to make a judgment call. We flag which is which, because the AI ones cost more to run and more to maintain.
Sales and lead workflows
1. Inbound lead enrichment and routing
The single most requested workflow we build. A lead comes in from a form, gets enriched, scored, and dropped into the right rep's queue.
Node map: Webhook trigger -> HTTP Request (enrichment API) -> Set (normalize fields) -> Switch (score by budget and location) -> CRM node (create or update contact) -> Slack (notify owner)
Breaks when: the enrichment API rate-limits you or returns a partial record. Without an error branch, the whole lead silently vanishes. Add a NoOp fallback path and a retry with backoff.
Verdict: Build it. This is the highest-ROI n8n example on the list for most teams.
2. Cold email reply classification
Inbound replies to outreach get read by an AI node, tagged as interested, not now, or unsubscribe, and routed accordingly.
Node map: IMAP trigger -> AI Agent (classify intent) -> Switch (route by label) -> CRM update -> conditional Slack alert for hot replies
Breaks when: the classifier sees an out-of-office auto-reply and marks it "interested." Add a pre-filter node that catches auto-responders before they hit the model.
Verdict: Build it, but keep a human review step for anything tagged interested until you trust the classifier.
3. Proposal follow-up sequence
After a proposal is sent, n8n waits, checks whether the document was opened, and nudges the rep or the prospect based on the answer.
Node map: CRM trigger (stage = proposal sent) -> Wait (48 hours) -> HTTP Request (check doc-open status) -> IF (opened?) -> branch to reminder email or rep task
Breaks when: the Wait node holds state across a workflow redeploy. On self-hosted n8n, a restart can drop in-flight waits. This is one reason how you host n8n matters more than people expect.
Verdict: Build it, but only on a stable host with execution data persisted.
Operations and back-office workflows
4. Invoice intake and approval
PDF invoices land in an inbox, get parsed, matched against a purchase order, and queued for approval over a threshold.
Node map: Email trigger -> Extract (PDF to text) -> AI Agent (pull vendor, amount, PO number) -> IF (amount over limit?) -> approval task or auto-post to accounting
Breaks when: a vendor changes their invoice layout. Structured extraction that relied on position falls apart. The AI node handles this better than regex, which is exactly why it earns its cost here.
Verdict: Build it. This is one of the few n8n examples where the AI node clearly pays for itself.
5. Inventory reorder alerts
Stock levels sync from the ERP, and low items trigger a draft purchase order to the right supplier.
Node map: Schedule trigger -> Database query (stock levels) -> Filter (below reorder point) -> Merge (supplier lookup) -> create draft PO -> Slack digest
Breaks when: the schedule fires during an ERP sync and reads a half-updated table. Pin the run to a quiet window and add a freshness check on the data timestamp.
Verdict: Build it. Rule-based, cheap, and it prevents expensive stockouts.
6. Onboarding automation
A signed contract kicks off account creation, welcome emails, a CRM card, and internal notifications, all in one run.
Node map: Webhook (contract signed) -> parallel branches: create app account, send welcome email, create CRM deal, post to Slack -> Merge -> log completion
Breaks when: one branch fails and the others succeed, leaving a half-onboarded customer. Wrap each branch so a failure is caught and reported rather than silently skipped.
Verdict: Build it, with per-branch error handling. Half-done onboarding is worse than none.
Support and content workflows
7. Support ticket triage
Inbound support email is classified, prioritized, and routed to the right queue with a suggested reply drafted.
Node map: Email trigger -> AI Agent (category and severity) -> Switch (route to queue) -> AI Agent (draft reply) -> create ticket with draft attached
Breaks when: the model over-escalates and marks everything urgent. Give it a rubric in the system prompt and cap how many tickets per hour can be flagged critical.
Verdict: Build the triage and routing. Keep the drafted reply as a suggestion, never an auto-send.
8. Knowledge base sync
When a help article is updated, the change flows into your vector store so the support AI stays current.
Node map: CMS webhook -> HTTP Request (fetch article) -> Set (chunk text) -> embeddings node -> upsert to vector database
Breaks when: an article is deleted but the old vectors linger, so the AI keeps citing dead content. Add a delete branch that removes stale embeddings by ID.
Verdict: Build it if you run a retrieval-based support agent. This is plumbing that a lot of AI automation projects forget until the answers go stale.
9. Content repurposing pipeline
A published post gets turned into a set of social snippets, reviewed, and scheduled.
Node map: RSS or webhook trigger -> HTTP Request (fetch full post) -> AI Agent (generate 5 snippets) -> approval step -> scheduler API
Breaks when: the AI node invents stats or quotes that were never in the source. Constrain it to the fetched text and keep the human approval gate.
Verdict: Build it, with review. Never let generated content auto-publish.
Data and reporting workflows
10. Daily metrics digest
Pull numbers from a few sources every morning and post a clean summary to Slack before standup.
Node map: Schedule trigger -> parallel HTTP Requests (analytics, CRM, billing) -> Merge -> Function (compute deltas) -> Slack message
Breaks when: one source is down and the Merge node waits forever. Set a timeout on each request and mark missing sources explicitly in the digest rather than failing the whole run.
Verdict: Build it. Cheap, rule-based, and it kills a recurring manual chore.
11. Form-to-spreadsheet with dedup
Form submissions land in a sheet, but duplicates get caught and merged instead of piling up.
Node map: Webhook -> Set (normalize email) -> lookup existing row -> IF (exists?) -> update or append
Breaks when: two submissions arrive within the same second and both read "not found" before either writes. Add a short queue or a unique constraint at the destination.
Verdict: Build it. This is a boring n8n example that quietly saves hours of cleanup.
12. Churn signal watcher
Usage data is scanned for drop-off patterns, and accounts at risk trigger a save play.
Node map: Schedule trigger -> database query (usage last 30 days) -> Function (flag decline) -> AI Agent (draft outreach angle) -> create CRM task for CSM
Breaks when: the decline rule is too sensitive and floods your team with false alarms. Tune the threshold on real historical churn before you turn it on.
Verdict: Build it once you have enough usage history to tune the signal. Not before.
The pattern behind all 12 n8n examples
Look across these workflows and the same three lessons repeat.
First, the trigger and the transform are easy. The failure handling is the actual work. Every verdict above hinges on what happens when a node fails, not when it succeeds.
Second, AI nodes earn their place only when the input is genuinely unstructured, like invoices with shifting layouts or free-text support emails. Everywhere else, a Switch node is cheaper, faster, and more predictable.
Third, hosting decides reliability. Wait nodes, in-flight executions, and retry state all depend on a stable environment. That is why we treat infrastructure as part of the build, not an afterthought.
Next step
Pick one workflow from this list, ideally a rule-based one, and build it end to end including the error branch. A single reliable automation beats a dozen half-finished experiments.
When you are ready to move past the easy wins into workflows that touch revenue and customer data, our workflow automation services can help you scope, build, and maintain them without the silent-failure surprises. Want the connective tissue between these and smarter decision-making? Our AI automation work is where the AI-node examples above graduate into full agents.
