Lead Generation Automation in n8n: A Real Build

We built this exact pipeline to find our own first clients, then ran it until it broke — twice. So this isn’t the happy-path version where you wire three nodes, screenshot the green checkmarks, and call it a system. This is the one that survived contact with real Google Maps data, missing websites, and an email API that lies to you politely.

By the end you’ll have a working lead generation automation that scrapes local businesses from Google Maps, finds their email, scores each lead with AI, writes a personalized first line, and drops the whole thing into a Google Sheet you can actually work from. Then we’ll spend equal time on the four things that quietly fail in production — because that part is the point.

What is lead generation automation? Lead generation automation is software that finds, enriches, and qualifies potential customers without manual work: a trigger fires, a scraper pulls business listings, an enrichment step adds contact details, and an AI step scores and personalizes outreach — so a person reviews a finished list instead of building it by hand.

If you’d rather we just build this for your niche, that’s literally our job — but you can absolutely ship it yourself with what’s below.

Not sure this is the right first automation for your business? We map that for clients in a free 30-minute automation audit — no pitch, a written recommendation.

What we’re building (the architecture)

Nine nodes, one clean path:

Schedule Trigger
  → Apify: Google Maps scraper       (find businesses)
  → Set: map fields                  (normalize the data)
  → Set: extract domain              (pull a clean domain from the website)
  → IF: has a website?
       → Hunter.io domain search     (find an email)
       → Set: pick best email
  → OpenAI: score + personalize      (qualify + write the first line)
  → Set: parse AI JSON
  → IF: skip empty rows              (guardrail)
  → Google Sheets: append/update     (your working lead list)

Tools and what each costs (estimates, June 2026):

Two honest notes before you build: Hunter only finds emails for businesses that have a website, and the cheap models are plenty for a one-line personalization — you do not need GPT-4o here.

Step 1 — Trigger and scrape Google Maps

Start with a Schedule Trigger (run it daily at 9am, or swap in a Manual Trigger while you’re testing).

Wire it into the Apify node using the Google Maps Scraper actor (compass/crawler-google-places). The only inputs that matter for a first run:

{
  "searchStringsArray": ["HVAC near New York"],
  "locationQuery": "USA, New York",
  "maxCrawledPlacesPerSearch": 25,
  "language": "en"
}

Keep maxCrawledPlacesPerSearch low (25) while testing — every place costs money, and you don’t need 500 rows to know whether the logic works.

Why Google Maps and not LinkedIn? For local-service niches (HVAC, plumbing, dentists, contractors) Maps is the highest-signal source you can scrape cleanly: it gives you phone, website, hours, and review count — and review count alone tells you who’s leaving money on the table.

Step 2 — Normalize the data and extract a domain

Add a Set node (“Map fields”) to rename the messy scraper output into clean fields you control: Lead Name, Company, Website, Email (default N/A), Source, Timestamp, Status (New Lead), and a Lead ID (use the domain or website so you can dedupe later).

Then a second Set node (“Extract domain”) to strip a website down to a bare domain Hunter can use:

{{ $json.Website && $json.Website !== 'N/A'
   ? $json.Website.replace(/^(?:https?://)?(?:www.)?/i, "").split('/')[0]
   : null }}

That regex turns https://www.cottamhvac.com/repair/ into cottamhvac.com. Small step, but Hunter fails silently if you hand it a full URL.

Step 3 — Branch on “has a website?”

Add an IF node: does domain exist and is it not empty?

• True → go find an email (Step 4).
• False → in the naive build, these get dropped. Hold that thought — it’s the first thing we fix in production.

Step 4 — Find the email with Hunter.io

Use the native Hunter node, operation Domain Search, with “Only Emails” on, authenticated with your Hunter credential. (You’ll see a lot of tutorials hardcode the API key into an HTTP Request URL. Don’t — that’s a leaked secret sitting in your workflow JSON forever. Use the credential.)

Then a Set node to pick the best result and keep it sane when nothing comes back:

Email:            {{ $json.data.emails.length ? $json.data.emails[0].value : 'N/A - Not found' }}
Email Confidence: {{ $json.data.emails.length ? $json.data.emails[0].confidence : 'N/A' }}

Step 5 — Score and personalize with AI

This is where a lead list becomes a lead pipeline. Add an OpenAI node (gpt-4o-mini) with a system prompt that does three jobs and returns strict JSON:

You are a lead qualification expert for HVAC businesses in New York, USA.
Tasks:
1. Lead Score (0-100): how good a fit is this business for our service?
2. AI Summary: 1-2 sentences on the business.
3. Personalized Message: one warm opening line referencing something specific.

Output JSON only:
{ "lead_score": number, "ai_summary": "string", "personalized_message": "string" }

Then a Set node to parse it back out:

Lead Score:           {{ JSON.parse($json.output[0].content[0].text).lead_score }}
AI Summary:           {{ JSON.parse($json.output[0].content[0].text).ai_summary }}
Personalized Message: {{ JSON.parse($json.output[0].content[0].text).personalized_message }}

Step 6 — Write to Google Sheets

Finish with a Google Sheets node (operation Append or Update), matching on Lead ID so re-runs update existing rows instead of creating duplicates. Map your clean fields to columns: Lead Name, Company, Website, Email, Lead Score, AI Summary, Personalized Message, Source, Timestamp, Status.

Add one last IF (“Skip empty rows”) before it so a half-empty scrape result never writes a blank row.

Run it. You now have a working lead generation automation.

This is the build we hand clients on day one. If you want it pointed at your niche and city — with the production fixes below already done — that’s a free automation audit away.

What breaks in production (the part nobody screenshots)

The build above works in a demo. Here’s what actually went wrong when we ran it for real, and how we hardened each one. These four fixes are the difference between a tutorial and a system.

1. Phone-only leads get silently dropped

A huge share of local businesses on Google Maps have no website — and our “has a website?” branch threw all of them away. For local service, those are often your best prospects (low digital maturity = more to fix). Fix: route no-website leads straight to the AI step with Email = N/A, score them on phone + reviews, and keep them for SMS/missed-call outreach instead of email.

2. The email API lies politely

Hunter returns N/A - Not found far more than you’d expect, and a low-confidence email is worse than none — it bounces and hurts your sender reputation. Fix: keep the confidence score, and don’t email anything under ~80 confidence. Treat those as call-only leads.

3. No error handling = one bad row kills the run

Out of the box, none of the nodes have error handling. One malformed scrape result or a Hunter rate-limit, and the whole execution stops — so you lose the 24 good leads because of 1 bad one. Fix: set retryOnFail on the Hunter and OpenAI nodes, and onError: continueRegularOutput so a single failure skips that row instead of the batch.

4. The AI confidently gets the facts wrong

Our first prompt was hardcoded to the wrong city — so the AI cheerfully described New York businesses as “located in Chicago.” It looked fine until you read it. Fix: never hardcode context the AI should read from the data; pull the city, name, and category from the actual row, and validate the JSON before you trust it.

That last one is why we say it on every build: AI is plumbing, not magic. The guardrails are the product.

Get the workflow

We packaged this whole pipeline as an importable n8n workflow — all nine nodes, pre-wired, with the production fixes above already in.

Import it (⋯ → Import from File), drop in your own Apify, Hunter, OpenAI, and Google credentials, and you’re running in about ten minutes.

Where this fits in your stack

A lead list is step one. The systems that actually win clients wire this into what happens next: the moment a high-score lead replies, it should trigger a follow-up sequence, a missed-call text-back, and a calendar link — automatically. That’s the same n8n logic, extended.

If you’ve made it this far, you’re not “researching automation” anymore — you’re deciding what to build first. That’s exactly the 30 minutes we spend in a free automation audit: we map your stack, find the one workflow with the fastest payback, and send you a written plan. No retainer, no pitch.

Related reading: n8n automation guide · How to automate lead generation · Build an AI agent in n8n · n8n vs Zapier