The first AI agent a brokerage should actually ship

Most brokerages do not have an AI problem. They have a speed-to-lead problem. A portal inquiry arrives while the agent is in a showing. The reply goes out that evening, if it goes out at all. The buyer already booked with the office that answered in four minutes.

The wrong first project is listing copy. Copy is visible and easy to demo. It is also optional. Intake is not. Every new buyer and seller file has to be read, scored, answered, and logged. That is a job with a beginning and an end.

A useful intake agent reads the inquiry, extracts contact, intent, market, budget, and timeline, then scores the file against your desk policy. It drafts a reply in your voice and stages a CRM record. It does not email the client until someone on your team clicks approve.

Policy is the product. Served markets, budget bands, investor rules, and quiet hours decide whether the lead is hot, held, or out of area. If those rules live only in a senior agent's head, the agent will guess. Write them down first.

Measure median first-response time, the share of inquiries with a complete file, and appointments booked from new leads over fourteen days. Include the minutes people spend reviewing drafts. An agent that drafts faster but creates a worse first impression is not a win.

Duplicates, out-of-area buyers, and incomplete 'just looking' notes will still happen. Route those to a clarifying template instead of a fake tour offer. The human checkpoint is where judgment belongs.

If you want this built around your CRM and your scripts, start with one office and one intake path. Expand after the numbers move.