One AI pipeline for every lead: $2.3M in new revenue for a real estate company in six months

TL;DR

Leads came in from the website, paid ads, referrals, and the team’s own outreach, and each source was handled a little differently. Some people got a call in five minutes, some in five days. Nobody could see every lead in one place or tell which sources were worth the money. We built one AI-powered pipeline that catches every lead, qualifies it, follows up on schedule, and hands it to an agent the moment the person is ready. With every lead treated the same way, 35% more of them turned into closed deals: $2.3M in new revenue in six months, at about 75% profit because the pipeline runs itself.

Context and users

• Agents wanted warm, ready-to-talk leads and hated chasing cold ones.
• The head broker wanted to know which of the four lead sources deserved budget.
• The leads themselves, people deciding whether to buy or sell a home, wanted a fast reply from a real person.

The problem, and how we found it

The broker’s ask was “more leads.” The assessment said the firm didn’t have a lead-volume problem; it had a lead-handling problem. We traced a sample of leads from each source to their outcome and found the same pattern four times:

• Response time depended on which channel the lead came through and who happened to see it.
• Follow-up depended on memory. Leads that didn’t answer the first time were often never contacted again.
• Attribution was guesswork, so ad spend couldn’t be judged.

Reframing “more leads” into “stop losing the leads you already pay for” changed the whole project.

Hypothesis

If every lead from every source entered one pipeline that qualified it, followed up on a fixed schedule, and handed it to an agent only when the person was ready to talk, the share of leads that closed would rise, agents would spend their time closing instead of chasing, and the broker could finally see which sources paid.

Scope, specs, and prioritization

Shipped in v1	Deferred
Capture from all four sources into one pipeline	New lead sources or ad campaigns
Qualifying questions to sort serious from curious	Automated pricing or listing recommendations
Follow-up sequence with fixed timing so nobody is forgotten	Post-close nurture (added later)
Hand-off to an agent when the lead is ready, with full context
One view of where every lead stands

Two specs I held firm on:

• Hand-off, not replacement. The AI never tried to sell a house. Its job ended the moment a person was ready to talk to an agent, and the agent got the whole conversation.
• Same treatment regardless of source. Tempting to prioritize paid leads; we didn’t, because the point was to find out which source actually converted when handled properly.

Key decisions and tradeoffs

• Fix handling before adding volume. The broker wanted more top-of-funnel. Choosing to fix the leak first meant a slower start and a much bigger result.
• Fixed follow-up cadence over agent discretion. Agents lost some freedom. In exchange, no lead went cold by accident.
• Measure closed deals, not replies. Reply rate would have looked great immediately. Closed deals took months to show but was the only number the broker actually cared about.

What shipped

Andres built the pipeline: capture from all four channels, qualification, scheduled follow-up, agent hand-off with context, and the shared status view. I ran discovery, the lead-tracing exercise, the scope and the “handling first” argument, and defined the closed-deal metric and how we’d measure it.

Metrics and outcomes

	Before	After (6 months)
Lead-to-closed-deal rate	Baseline	+35% (relative)
New revenue attributable to the pipeline		$2.3M
Profit share of that revenue		~75%
Lead sources handled consistently	0 of 4	4 of 4
Visibility into where each lead stands	Group chats and memory	One shared view

What I’d do differently

• Lock the attribution model before launch. We agreed on “closed deals” as the metric but spent time later arguing about which source got credit for a lead that touched two channels.
• Tune the qualifying questions with agents sooner. The first version asked one question too many, and a few good leads dropped before the agents caught it.
• Report weekly from day one. The broker saw the big number at six months; smaller weekly numbers would have built confidence earlier and caught the qualifying-question issue faster.

My role vs. the team

I owned discovery, the reframing from volume to handling, scope and sequencing, the metric definition, and the broker relationship. Andres owned the build across all four channels. The head broker owned agent adoption, which is why the hand-off rules stuck.