A real estate agency was losing deals every week to faster competitors. We built a fully automated lead qualification and response system using WhatsApp Business API, n8n, and the Claude API — and response time dropped from an average of 6 hours to under 90 seconds.
Thinkiyo Studio
January 15, 2026 · 6 min read
When a mid-sized real estate agency in Brisbane came to us, their problem was simple to state and painful to live with: leads were coming in around the clock from six different sources — web forms, Facebook Lead Ads, REA Group, Domain, Instagram DMs, and cold inbound calls — and their team of eight agents was responding to them in batches, usually first thing in the morning.
The average time between a lead submitting their details and an agent calling them back was 6 hours and 14 minutes.
In real estate, that is an eternity. A motivated buyer or seller who submits an inquiry at 7 PM on a Tuesday has usually spoken to two other agents by the time anyone calls them back at 9 AM Wednesday.
Here is exactly how we fixed it.
Before we built anything, we mapped the existing workflow. It looked like this:
The team was not lazy. They were just operating a process built for a slower era. Every step was manual, every handoff created delay, and there was no visibility into which leads had been contacted and which hadn't.
The result: they were converting roughly 3.2% of inbound leads to booked appraisals. Industry average for well-run agencies with fast response is closer to 11–14%.
We built a three-stage automation pipeline. Here's what it does end-to-end.
The first challenge was aggregating leads from six sources into a single processing queue. We used n8n's webhook nodes and native integrations to capture leads from:
Each lead, regardless of source, gets normalised into a consistent data structure: { name, phone, email, source, propertyInterest, budget, timestamp }. Incomplete records get flagged for enrichment rather than discarded.
Once a lead is normalised, it is passed to a Claude API call with a structured prompt that evaluates qualification signals:
Claude returns a structured JSON object:
{
"tier": "hot",
"score": 87,
"summary": "Active buyer with 30-day timeline, budget aligns with property, mobile confirmed",
"recommendedAction": "immediate_whatsapp",
"personalisedMessage": "Hi [name], thanks for your interest in [property]. I'd love to answer any questions..."
}
The prompt was refined over about three weeks of A/B testing against leads the agents had manually qualified. We achieved 89% agreement between Claude's tier assignment and the agents' retrospective assessment.
Based on the qualification tier:
Simultaneously, the n8n workflow:
We used the Meta WhatsApp Business API accessed via a verified Business Manager account. The agency's number was registered as a WhatsApp Business number, and we set up approved message templates for the initial outreach (required by Meta for the first message to a new contact).
Once a lead replies, the conversation moves to session messaging, which has no template restrictions. We built a basic intent-detection layer in n8n that listens for replies and routes them: if someone replies "yes" or books via Calendly, the agent is notified. If they ask a question, it goes to a holding queue for agent pickup within 2 hours.
| Metric | Before | After |
|---|---|---|
| Average lead response time | 6h 14m | 88 seconds |
| Lead-to-appraisal conversion | 3.2% | 9.1% |
| Agent time on lead admin | ~2.5 hrs/day | ~20 min/day |
| Leads contacted outside business hours | 0% | 100% |
| Appraisals booked per month | 18 | 47 |
The agency added two new agent hires within six months — not because the automation failed, but because they had more qualified opportunities than they could service.
One thing we underestimated: the importance of the human handoff design. The first version of the system was too aggressive — hot leads received a WhatsApp message AND an automated follow-up call (via a VoIP auto-dialler) within 10 minutes. Several leads found this intrusive and complained.
We dialled back the auto-call, giving leads the option to schedule via Calendly instead of calling them directly. Conversion actually improved — people respond better when they feel in control of the timing.
The second lesson: invest in the evaluation harness early. We spent three weeks tuning the Claude qualification prompt, and that work paid off. But we should have built the evaluation dataset (manually labelled leads) before writing a single line of n8n code.
This architecture works best when:
If you are working from spreadsheets or a single shared inbox, the first step is getting a CRM in place — we can help scope that too.
The total build took four weeks and has paid for itself many times over in recovered revenue.
Share this article
Zapier is the easiest to start with, Make gives you more power at lower cost, and n8n lets you run everything yourself. But the right answer depends on your technical team, your data sensitivity requirements, and how complex your workflows actually need to be.
Read more →GuideWhatsApp has over 2 billion active users globally and strong adoption across Australian consumer demographics. Here is how to get API access, build compliant automated flows, and design a human handoff that keeps conversations feeling human.
Read more →20-minute call. No pitch deck. Just a direct look at where automation ships ROI fastest.