7 n8n Workflow Examples for Lead Generation

Most n8n lead generation content is a screenshot of a form connected to a Slack message. That is a demo. A real workflow survives a rate limit, a duplicate submission, a vendor that returns an empty body with a 200 status, and a rep who is on holiday.

Below are seven n8n workflow examples lead generation teams run, all of which I have built or rebuilt on client systems, each written as a trigger, a node chain, and the specific failure mode that will bite you. If you have never wired a node in n8n, read the n8n tutorial for beginners first, then come back and build these in order.

What are the best n8n workflow examples for lead generation?

The best n8n workflow examples for lead generation are the seven listed here: contact enrichment, ICP scoring, inbound routing, LinkedIn-to-CRM capture, review monitoring, churn and expansion signals, and dead-pipeline re-engagement. Together they cover the whole path from an anonymous contact to a task sitting in a rep's queue.

They all share one shape: a trigger, a deduplication step, an enrichment or scoring step, a CRM write, and an error branch. Learn that shape once and the seven stop being seven builds and become variations on a single pattern. I use exactly this skeleton on the n8n automation builds I ship for sales teams.

"A lead generation workflow without a deduplication step is a machine for annoying your best prospects at scale."

Workflows that turn a raw contact into a qualified lead

The first two workflows do the work a junior SDR used to do by hand: fill in what the form did not ask for, then decide whether the lead is worth a human. Build them as separate workflows, not one, so a broken enrichment vendor does not stop scoring.

1. Form-to-enrichment

Trigger is a Webhook node on your form endpoint. Chain: Webhook, then Remove Duplicates keyed on the lowercased email, then an HTTP Request to your enrichment provider, then Edit Fields to flatten the response, then an upsert into the CRM, then a Slack message to the owning channel.

Failure mode: enrichment APIs frequently return a 200 with a null or partial body for unknown domains. If you map fields blindly, you write empty company names and null headcounts over data a rep entered by hand. Put an IF node after the HTTP Request that checks one required field, and route unknowns into a manual queue instead of into the CRM.

2. ICP scoring with an LLM

Trigger is a Schedule Trigger every fifteen minutes over records created since the last run. Chain: query new records, a Code node that builds one compact profile string, an OpenAI or Anthropic Claude node returning strict JSON, a Switch node that bands the score, and a write-back of both the score and the reasons.

The model is a classifier, not a judge. It gets firmographics, the stated use case and the traffic source. It does not get deal value, and it does not get the rep's name. That is the same discipline behind ProLeads, where a natural-language description of a buyer resolves to a filtered list rather than a vibe.

// Code node: validate the model output before it touches the CRM
const out = [];

for (const item of $input.all()) {
  let parsed;
  try {
    parsed = JSON.parse(item.json.message.content);
  } catch (err) {
    throw new Error('scorer returned non-JSON for ' + item.json.email);
  }

  const score = Number(parsed.score);
  if (!Number.isInteger(score) || score < 0 || score > 100) {
    throw new Error('score out of range for ' + item.json.email);
  }

  out.push({
    json: {
      email: item.json.email,
      score,
      reasons: Array.isArray(parsed.reasons) ? parsed.reasons.slice(0, 3) : [],
      promptVersion: 'icp-v4',
      scoredAt: $now.toISO(),
    },
  });
}

return out;

Failure mode: score drift nobody can explain. The prompt gets edited in March, and in May a sales manager asks why a lead scored 82. Store the prompt version alongside every score, as above, and treat a prompt change like a schema migration.

Workflows that get the lead to the right human fast

Speed to first touch is the one lead generation variable automation controls completely. These two workflows exist to remove the gap between a lead arriving and a named person owning it.

3. Inbound routing and round-robin

Trigger is the same Webhook, or a mailbox trigger for inbound email. Chain: classify intent with a Switch node on keywords or a small LLM call, look up the next owner, create the CRM task, send a Slack direct message with the response deadline, then a Wait node, then an escalation branch if the record is still untouched.

Failure mode: round-robin state. If you keep the last-assigned index in workflow static data or a variable, it resets on restart or drifts when two executions overlap, and one rep quietly receives forty leads. Keep the counter in Postgres or Supabase and increment it with an atomic update inside the query. I rebuilt exactly this on a 24-stage awards funnel CRM after ownership had quietly gone lopsided for weeks.

4. LinkedIn-to-CRM capture

Trigger is a Webhook fed by a browser action you take deliberately, or a file-drop trigger watching an export you are entitled to download. Chain: normalise name and company, dedupe on company domain, enrich by domain, upsert into the CRM, then add to a sequence only after a human confirms.

Failure mode: this is the workflow people build on unofficial scrapers, and it is the one that fails hardest. LinkedIn's User Agreement prohibits automated scraping and unauthorised software access, so the tooling breaks with every UI change and the account carries the risk. If you need public web data at volume, take it from sources that permit it, which is the whole argument in scraping Google search results with n8n legally.

Workflows that mine leads out of signals you already have

The last three workflows generate pipeline without buying a single list. They read public complaints, your own product usage, and your own dead deals.

5. Review and mention monitoring

Trigger is an hourly Schedule Trigger. Chain: HTTP Request or RSS pulls of review feeds and forum searches, then the Remove Duplicates node in its previous-executions mode, then an LLM classification of sentiment and intent, then a Switch that turns competitor complaints into leads and your own complaints into support tickets.

Failure mode: duplicate storms. Feeds re-publish, and unstable IDs make every poll look like new content. n8n's documentation for the Remove Duplicates node describes a mode that removes items processed in previous executions and keeps a deduplication history, so key it on the most stable field available, usually a permalink URL rather than a title.

6. Churn and expansion signals

Trigger is a nightly Schedule Trigger. Chain: a SQL query against product usage in Postgres or Supabase, a Code node that computes the change against the prior period, a Switch that sends sharp drops to a customer success task and sustained growth to an expansion lead, then a CRM write. For SaaS teams, this is usually the highest-yield workflow on the list.

Failure mode: silent query drift, and it is the longest-hiding bug I have seen in this category. A renamed column, a changed enum or a timezone boundary makes the query return zero rows. Zero rows is not an error, so n8n reports a successful execution every night for a month. Add a guard node that throws when the row count is zero on two consecutive runs.

7. Dead-pipeline re-engagement

Trigger is a weekly Schedule Trigger. Chain: query closed-lost deals and contacts with no activity for ninety days, filter out anyone with an open opportunity, have an LLM draft one specific reason to reach out from the last call note, push the drafts into a Slack approval step, send only what a human approved, then log the outcome back to the CRM.

Failure mode: contacting people you are not allowed to contact. Suppression has to be a hard filter early and a second check immediately before send, because unsubscribes land between the two steps. Keep the human in the loop. An unsupervised weekly re-engagement blast is how a sending domain gets its reputation destroyed.

How do you stop n8n lead generation workflows from breaking?

You harden them the same way every time: make every write idempotent, make every silence loud, and never let one vendor outage take down the chain. Seven steps cover most of it.

Which of these should you build first?

Build routing first, then enrichment, then scoring. Routing pays back immediately because it removes dead time between a lead arriving and a human touching it, and it needs no external vendor. Scoring is last because a score is worthless until the routing it feeds is trustworthy.

Some things should never live in n8n. High-volume scraping, anything with a per-record latency budget under a second, and stateful multi-step processes with retries and back-off belong in a real service. The distributed scraping platform I run at over 100K jobs a day is Python and queues, not a canvas. n8n orchestrates it; n8n does not do it. That split is the core of how I scope AI automation work.

What does it cost to run these workflows?

Cost has four components: the n8n plan or your hosting bill, execution volume, enrichment credits per lookup, and LLM tokens per scored record. Only the first is fixed; the other three scale with lead volume.

Vendor pricing moves constantly, so read n8n's official pricing page for current tiers rather than any figure in a blog post. If the execution bill is what is pushing you toward a server, weigh the real operating cost first in self-hosting n8n: costs and trade-offs.