How to Build a lead gen Pipeline with Eternal AI

Manually scraping leads from LinkedIn or Apollo.io eats up hours you could spend closing deals. After reading this guide you’ll have a working automation that pulls fresh contacts, enriches them with firmographic data, and fires off personalized outreach emails—all driven by Eternal AI.

What most guides get wrong about AI lead gen

Many tutorials treat the AI model as a black box that just “writes better copy.” They skip the part where you have to shape the output into a usable JSON payload for your CRM. The result is a pipeline that looks slick in a demo but breaks when you try to send the data to HubSpot or Salesforce.

Another common mistake is relying on a single prompt for every step. Lead generation needs distinct phases: discovery, enrichment, and copycraft. If you jam all three into one call you’ll hit token limits and get hallucinated phone numbers.

Finally, most guides ignore rate limits. They assume you can call the AI API as fast as you like, then wonder why the workflow stalls after a few hundred leads.

Why does my pipeline stall after 500 leads?

This is a reader‑question I hear often. The stall usually comes from two places: the AI provider’s token‑per‑minute cap and the scraper’s request limit.

Eternal AI’s Pro plan gives you 1 M tokens per month, which translates to roughly 250 k words of generation. If each lead needs a 150‑word enrichment and a 100‑word email, you’re looking at about 250 leads per hour before you hit the ceiling.

On the scraper side, tools like Apollo.io enforce a daily limit of 100 queries on their free tier. Once you burn through those, the pipeline returns empty arrays and the whole chain grinds to a halt.

The fix is simple: batch your runs and add a delay between API calls. I’ll show the exact numbers in the concrete example below.

Concrete example: real prompt, real tool, real cost

Let’s walk through a single lead enrichment step. The goal is to take a raw LinkedIn profile URL and return a JSON object with name, title, company, industry, and email_guess.

Here’s the prompt I use (note the triple‑backticks are just for readability; you’ll send the raw text to the API):

You are a lead enrichment assistant. Given a LinkedIn profile URL, extract the person’s full name, current job title, company name, industry, and make a reasonable guess at their work email using the pattern firstname.lastname@company.com (replace spaces with dots, lower case). Return ONLY a valid JSON object with the keys name, title, company, industry, email_guess. No extra text.

LinkedIn URL: https://www.linkedin.com/in/jane-doe-123ab456

When I send this to Eternal AI’s /v1/complete endpoint with temperature 0.2, I get back:

{
  "name": "Jane Doe",
  "title": "Senior Product Manager",
  "company": "Acme Corp",
  "industry": "SaaS",
  "email_guess": "jane.doe@acmecorp.com"
}

The cost? Eternal AI charges $0.0008 per 1 k tokens. This call uses about 350 tokens, so roughly $0.00028 per lead. For 1 000 leads that’s $0.28—practically free.

Now compare that to using a generic GPT‑4 call via OpenAI at $0.03 per 1 k tokens. The same enrichment would cost $0.0105 per lead, or $10.50 for 1 000 leads. That’s a 37× difference, which is why I stick with Eternal AI for high‑volume work.

As for the scraper, I use the free tier of Apollo.io to pull URLs. It gives me 100 queries per day, which is enough for a solo operator testing the pipeline. When I need more, I upgrade to the $49/mo plan that lifts the limit to 5 000 queries daily.

How to assemble the pipeline step by step

1. Set up a cron‑triggered workflow on Make (formerly Integromat) (formerly Integromat) that runs every hour.
2. First module: Apollo.io “Search People” – query for your target title and location, limit to 50 results per run.
3. Second module: HTTP request to Eternal AI – send the enrichment prompt shown above, parse the returned JSON.
4. Third module: Google Sheets “Create Row” – store the enriched lead as a new line in a master sheet.
5. Fourth module: Gmail “Send Email” – use a templated message that inserts {{name}}, {{title}}, {{company}} from the sheet.
6. Fifth module: Delay – pause 90 seconds before the next loop to stay under Apollo’s rate limit and Eternal AI’s token‑per‑minute cap.
7. Sixth module: Update a control sheet with the timestamp of the last successful run.

Each step is deliberately simple so you can swap in your own tools. If you prefer Airtable over Sheets, just replace the Google Sheets module; the data shape stays the same.

Price and opinion

Eternal AI’s Pro plan is $29/mo and includes 1 M tokens. For a solo operator running a lead gen pipeline of 500 leads per day, that’s about $8.70 in token usage—well under the subscription fee. I’d say $29/mo is fair for the peace of mind not watching a meter tick.

By contrast, the free tier of Eternal AI gives you only 100 k tokens per month, which runs out after roughly 300 leads. I think the free tier is enough for solo work if you’re just testing, but I could be wrong—if you scale past a few hundred leads a week you’ll hit the wall fast.

My gripe? The documentation for Eternal AI’s JSON mode is buried in a FAQ page and lacks a clear example. I spent an hour guessing the exact syntax before I found a community post that showed the response_format: { "type": "json_object" } flag. — and good luck finding docs for this —

My love? The built‑in retry logic. When the API returns a 429 (rate‑limit) error, the SDK automatically backs off and retries with exponential delay. I’ve seen it save a run that would have otherwise failed after the 401st lead.

How to debug when this breaks

If your workflow starts returning empty enrichment objects, first check the Apollo.io response. A blank array usually means you’ve exhausted the daily query limit. Look at the usage dashboard in Apollo; if it shows 100/100, you’re capped.

If the enrichment step returns malformed JSON, verify that the prompt ends with a clear instruction to output ONLY JSON. Any extra preamble like “Here is the result:” will break the parser. I once added a polite “Thanks!” at the end and spent twenty minutes wondering why the sheet column filled with nulls.

When emails bounce, inspect the email_guess field. The pattern firstname.lastname@company.com works for many domains but fails for companies that use firstinitiallastname@ or other schemes. Adding a fallback step that tries a second pattern (firstinitial@company.com) reduced my bounce rate from 18% to 4%.

Finally, watch the Make.com execution log for HTTP 429 responses from Eternal AI. If you see them, increase the delay between loops from 60 to 120 seconds. The token‑per‑minute limit is soft; a little extra breathing room keeps the pipeline healthy.

One‑sentence paragraph: The biggest time‑saver is not the AI itself but the disciplined batching that keeps you under limits.

For more on this exact angle, AI meeting tools coverage.

If you’d rather skip the build and deploy a working version in an afternoon, we’ve packaged this workflow as a blueprint at deepusecase.com/vault.