We recently shipped a voice AI and SMS automation system for a real estate lead generation company. The goal was simple. Every buyer lead gets called within 2 to 5 minutes of form submission, qualified, and logged automatically.

Here is how we built it and what we learned.

The Business Problem

LeadstoClosings generates buyer leads through Meta, Google, and TikTok ads and sells them to real estate agents across the US.

The operational gap was straightforward. Agents were not calling leads fast enough. Contact rate drops 400% after 5 minutes. Most agents make 1 to 2 attempts. Converting a lead takes 5 to 8. That math never worked in anyone's favor.

On top of that, agents would request lead replacements claiming leads were unreachable. With no call logs or timestamps, our client had no way to verify that. Every replacement issued was direct margin loss.

Lead generated: $35
Lead sold: $75
Margin after replacements: shrinking fast

What We Built

A fully automated AI voice and SMS qualification pipeline that runs the moment a lead enters the CRM.

The flow looks like this:

Lead submits form
       ↓
GoHighLevel CRM receives lead
       ↓
Make.com triggers automation
       ↓
VAPI places AI voice call
       ↓
ElevenLabs voice runs conversation
       ↓
Call outcome returned to Make.com
       ↓
GoHighLevel CRM updated
       ↓
SMS sent if no answer

Tech Stack

GoHighLevel handles CRM, pipelines, workflows, and automation triggers.

VAPI manages the AI voice conversation, call control, and speech processing.

ElevenLabs provides the realistic voice output for the AI agent named Sarah.

Make.com acts as the integration layer connecting all three systems together.

The Voice Agent

The AI agent Sarah calls every new lead within 2 to 5 minutes. The call stays under 3 minutes and covers three things only.

Is the person still interested in buying a home. Have they signed a buyer representation agreement with another agent. When are they available to speak with a partner agent.

The qualification logic is deterministic. No fuzzy scoring. No guessing.
A lead is qualified if they confirm they are still buying and have not signed with another agent.

A lead is not qualified if they are not buying, already signed with someone, or request no contact.

// Simplified qualification logic
function qualifyLead(response) {
  const { stillBuying, signedWithAgent, requestNoContact } = response;

  if (requestNoContact) return "do_not_contact";
  if (!stillBuying) return "not_interested";
  if (signedWithAgent) return "already_with_agent";
  if (stillBuying && !signedWithAgent) return "qualified";

  return "no_answer";
}

CRM Outcomes

After every call the CRM pipeline updates to one of five states.
Qualified. Not interested. Already working with agent. Do not contact. No answer.

SMS Fallback

If the lead does not answer, Make.com triggers an SMS immediately through GoHighLevel.

// Make.com webhook payload to trigger SMS
{
  "contact_id": "{{lead.id}}",
  "message": "Hi, this is Sarah from Home Front Connect. 
  You recently requested information about buying a home. 
  When would be a good time for one of our partner agents 
  to reach you?",
  "trigger": "no_answer"
}

What Phase 2 Looks Like

The client is already planning the next build. If a buyer answers and qualifies, the AI will transfer the call live to the assigned agent. Routing will use CRM tags and conditional logic. Every transfer attempt and agent response will be logged.

Build Time

Phase 1 took roughly 30 hours total covering architecture design, prompt engineering, voice system integration, CRM integration, automation workflows, and testing.

Key Takeaways

Speed is the entire point. Getting to a lead in 2 minutes versus the next morning is the difference between a conversation and a cold call to someone who forgot they clicked an ad.

Deterministic qualification logic matters more than smart AI here. Strict rules produce consistent outcomes. Fuzzy logic produces inconsistent lead quality that agents will complain about.

CRM traceability solves business problems beyond automation. The call logs alone changed how our client handles replacement requests.

This same architecture works for mortgage leads, insurance leads, solar leads, or any business selling time-sensitive inbound inquiries.

Happy to answer questions on the VAPI setup or the Make.com workflow in the comments.