AI doesn't start with technology. It starts with data.

Fábio, one of the founders of CV CRM, invited me to episode 12, season 4, of CV na Prática, their podcast for the Brazilian real estate market. The topic was what changes when real estate enters the era of AI agents. Here is what I argued, for anyone who doesn't have the 32 minutes. The episode is in Portuguese.

Watch the full episode on YouTube.

Every piece of data comes from somewhere

Before Tailor.ia, I helped build 123i, a property portal. To have reliable information, we catalogued 38,000 buildings by going in person to Brazil's 40 largest cities, taking photos and filling in spec sheets by hand. I tell this story because the lesson still holds: good AI depends on good data, and good data takes work.

When there is no API or integration, I say start with a spreadsheet. Export whatever you have, load it into a database and go from zero to one. Buying a tool doesn't solve the problem. Working on the tool's data does.

From lead to after-sales, without the broken telephone

A well-built AI agent follows the whole journey:

• A lead becomes a deal. The lead arrives from a campaign, a portal or a QR code on a construction hoarding. The AI works out whether the interest is real and merges several contacts from the same person into a single deal.
• The visit changes. "I can't make it today, my daughter is sick, let's try Saturday." The agent reschedules, tells the sales agent and already prepares alternatives in case the property doesn't win the buyer over.
• Paperwork. It asks for ID, proof of address and the other documents the mortgage needs, tracks what is missing and keeps the sales agent posted, so nobody has to chase papers.
• After-sales. A copy of a payment slip, the apartment floor plan, opening a complaint. It's half a dozen repetitive requests. And it can spot patterns, like a floor tile that starts failing in several units of the same building.

The result is that the sales agent becomes a closer: they reach the bottom of the funnel with a customer who has already been served, informed and organized.

Tailor-made, not off the shelf

Our AI has a base layer, the rice and beans, as we say in Brazil. But the real work is tailor-made, which is where the name Tailor.ia comes from. Every company has its own rules, rhythm and vocabulary, so we learn the business, wear the client's jersey and help run the operation. Delivering just an AI is delivering more of the same. What I want to deliver is results.

One of our clients more than doubled their sales volume, but it took at least a year of work on process and data. Our newest clients have been with us for eight years, because this is work built on trust and day-to-day effort.

Architecture and governance

Every AI has outages: Claude, OpenAI and Gemini all go down now and then. So our architecture has a selector that picks which model to use based on how stable each one is. The client never notices the switch.

On governance, each client has a dedicated environment on servers in Brazil, and no client can see another's data. We don't use personal data to train anything, in line with Brazil's data protection law (LGPD). Inside the agent, tasks are split: one part identifies the product, another fetches the information, and a reviewer checks the answer before it goes out.

What comes next

The next step is making systems talk to each other: the CRM with its AI, our AI with theirs, through APIs and MCP. The question that matters isn't how many leads came in today. It's how to sell more, in less time, with a better experience for the customer.

If you want to bring this conversation to your team, it's exactly what I do in the AI Team Immersion and in my talks.