Michel Moubarak

We entered an incubator with three ideas, did the customer development, research and SWOT analysis, and came out with RentHunt: a swipe-based matching app for landlords and renters.

Then we launched it.

Almost nobody downloaded it.

We were running out of money and needed to find out whether the problem was real before building anything else. So we stopped building. We ran a Facebook ad telling people looking for an apartment to text us. When they did, we manually searched for homes and texted results back.

People responded.

The experiment showed us that searching for a home was painful enough that people wanted help. We eventually shifted from rentals toward buyers, where agents saw more value, and the repeated conversations gave us something worth automating.

That eventually became Roof AI.

Those early SMS conversations taught us something else.

People didn’t naturally search for homes using filters and dropdowns. They said things like, “I need a 3.5 near Peel for July 1.”

So in 2016 we built natural-language home search into brokerage websites.

It became a major wow moment in demos. But online, almost nobody knew what “natural-language search” meant or why they should care.

The technology was ahead of the category.

So I started creating content and thought leadership to explain conversational search, changing consumer behavior and what this could mean for real estate.

Seven years later, Zillow launched natural-language home search.

I was selling the idea that brokerages were losing opportunities because they responded too slowly to online leads.

There were plenty of industry statistics supporting the problem.

They weren’t landing.

So I secret-shopped 74 of the largest US brokerages. I submitted an inquiry through every website and measured what happened.

41% never responded. Only seven responded within five minutes.

Then I emailed each brokerage its own response time.

The research became sales material, industry education and something I could use for years to make an abstract problem personal to the person sitting across from me.

Years later, industry analyst Mike DelPrete ran a similar experiment and found the problem was still there.

Our first US customer had a strong internal champion. She believed in Roof AI and pushed the product internally.

We launched.

Then the CEO saw the chatbot on a luxury page and wanted it removed. His objection wasn’t performance. He believed luxury real estate should remain human-first.

I had never spoken to him.

After that, I changed how I sold enterprise software.

I deliberately asked champions to bring senior stakeholders into the process before launch. When they couldn’t join, we created material our champion could use internally.

The objective wasn’t simply to win a champion anymore. It was to understand who else could influence the decision and make sure they understood what we were doing.

One of our early HomeServices customers loved what we were doing.

We kept her close, involved her in product discussions and treated her almost like an informal advisor. Eventually, she introduced us to HomeServices of America.

We signed an MSA.

At first, that felt like the big win.

It wasn’t.

The agreement gave us credibility, preferred pricing and permission to approach the network. It generated no revenue by itself. We still had to sell each brokerage individually.

We cold outreached. We went to conferences. We built strong customer relationships. And after successful implementations, we systematically asked customers a simple question:

“Who else do you think would find this interesting?”

Eventually referrals and word of mouth started doing much of the work.

More than 20 HomeServices brokerages became customers.

We thought growing Roof AI meant graduating from founder-led sales.

So we hired a VP of Sales.

I built the RevOps infrastructure around the role: funnel definitions, dashboards, activity metrics and sales enablement.

But the underlying sale wasn’t ready to be handed off.

Selling Roof AI required product knowledge, real-estate knowledge, consulting, credibility and flexibility. Enterprise customers also valued having direct access to the founders.

We eventually returned to founder-led sales.

The RevOps infrastructure stayed.

If I were building the function again, I would sequence it differently: founder sells, an SDR helps fill the founder’s calendar, the outbound motion gets proven, then an AE takes over more of the closing. Sales leadership comes when there’s actually a repeatable sales function to lead.

Roof AI still relied heavily on scripted decision trees when ChatGPT arrived.

Then something started happening on my monthly customer calls.

Clients kept asking what we thought about ChatGPT and how they could use it. They also talked about how much smoother and more natural those conversations felt.

The benchmark had changed. Our customers weren’t comparing Roof AI only with other real-estate technology anymore. They were comparing the experience with the AI they were using themselves.

I brought that signal back into the company. It became the catalyst for moving Roof AI toward large language models.

I stayed involved through product meetings, testing, customer requirements, prioritization and feedback.

Then I worked across the commercial launch: positioning, messaging, PR, sales material, client communications, onboarding and training.

We went further for customers too. We created presentations, marketing material and videos they could use to launch the technology to their own agents and customers.

Eventually every customer moved to the new model.

Roof AI has an unusual customer-success problem.

It’s B2B2C.

Our customers pay for the product, but their consumers interact with it. That means Roof AI can quietly do its job every day while the person approving the renewal barely experiences it.

I learned that creating value wasn’t enough. We had to make customers feel that value.

For years, I did this manually through recurring calls, analysis and reports.

Eventually I worked with foundation models to review the conversations happening across customer websites and turn them into useful business intelligence.

Instead of simply reporting “conversations” or “leads,” we could show customers what consumers were actually asking, what percentage of questions were resolved automatically, when demand appeared, what information was missing from their data, and where new opportunities existed.

For one brokerage, for example, the analysis showed that 51% of conversations happened outside office hours and two-thirds of questions were answered instantly. It also identified specific gaps in their listing data that prevented the AI from answering customer questions.

The reporting itself became part of the value.