An AI agent for real-estate agencies, from the inside

Barceloneta Premium, a Barcelona agency, received dozens of WhatsApp messages a day from people interested in renting. Each inquiry demanded five to ten minutes of manual checking before anyone knew whether it was worth pursuing. Multiply those minutes by dozens of messages and you get the bill that was eating the team, a full working day scattered across ten-minute interruptions.

The obvious answer would have been hiring someone to do the triage. The interesting answer was removing the triage altogether.

What the agent does, step by step

The agent receives each message and holds the conversation needed to extract three things, the reason for the inquiry, the budget and the documentation available. It does not push a form at people. It asks the way a team member would ask, in the language of the person writing.

With the information complete, the agent neither decides alone nor leaves the conversation hanging. It sends the team an email summary with a verdict, suitable or not suitable, and a paragraph justifying why. The person at the agency opens the email, reads three lines and decides whether to schedule the visit. The judgment stays theirs. What disappears is the ten minutes of checking it used to take to reach that judgment.

The result, in hours rather than promises

More than three hours a day saved on handling incoming requests alone. The team went from triaging to scheduling visits, which is the part of the job that produces revenue. And interested people get an immediate reply at any hour, including the rental-demand peaks where they used to cool off waiting.

That is also the yardstick we use to decide whether an AI project deserves to exist, that it moves a business figure and not a feeling. We tell that story in measure AI by profit.

What it takes for it not to be a demo

An agent like this touches personal data, calendars and reputation, so the invisible part weighs more than the visible one. The conversation leaves a complete trail, every request is recorded with its verdict and its justification, and the filter that decides what gets escalated does not live in the model’s instructions. It lives in code that always runs. These are the same construction rules we apply to all our conversational AI work, whether it serves an agency or a clinic.

And after rentals, the rest

The signal that a system works is what happens next. The agency is extending the agent to property sales and to internal processes, on the same base that already filters rentals. Starting with one painful process and growing from there is the pattern we see repeat in the projects that survive.

The maths we did here you can do this afternoon. How many minutes does each incoming inquiry cost, and how many arrive per day? Multiply. You are already paying that number, just without an invoice.

If your agency looks like this, the AI for real estate vertical explains the full system with its proof in production. And if you are in another sector with the same bottleneck, the conversation starts the same way, telling us where the hours go.