Section 1
Speed to first contact decides the deal
Property enquiries behave like any other high-intent lead. The buyer is looking at several properties on the same evening and the agent who responds first has an advantage that has nothing to do with skill. Yet most brokerages route enquiries into an inbox that gets read the following morning. Automating that response is not a chatbot project. It is qualification and routing: read the enquiry, match it against the listing and the buyer's stated position, answer the questions the listing already answers, and put a booked viewing slot in front of them. The agent's time goes to the conversations that need an agent. Fraud checks on the transaction side are covered in [Using AI for Real-Time Fraud Detection](/blog/using-ai-for-real-time-fraud-detection).
Section 2
The three workflows, ranked by payback
Transactions pay best. A sale involves a long chain of documents, deadlines, and third parties, and progress stalls silently. Automated chasing, document classification, and a status view that updates itself removes the coordination work that currently fills an administrator's week. Viewings pay next. Scheduling, confirmations, reschedules, access instructions, and the follow-up note afterwards are all rule-bound and all currently done by phone. Listings pay least, but they are the easiest to start with, and that is a legitimate reason to begin there while you build confidence. Just do not mistake a faster description for a better business. The description was never the constraint.
Section 3
Where an agent must still hold the pen
Three things stay human in property. Anything that constitutes advice on price or value. Anything that goes into a contract or a legally significant disclosure. Anything that discriminates between buyers, where automated screening on the wrong attribute creates exposure that is both legal and reputational. Automate the coordination around those decisions, never the decisions themselves.
Section 4
Starting with one transaction workflow
Pick a single stage, such as the period between offer accepted and exchange, and measure it honestly. How many days does it take, how many of those days are waiting on a document, and how many transactions fall through in that window. Then automate the chase, not the judgment. The system tracks which documents are outstanding, requests them, files them when they arrive, and escalates when a deadline approaches. A person still reviews anything a client sees. Run it inside the CRM the agents already use, because a separate transaction dashboard is a system that gets updated for a fortnight and then stops.
Section 5
Client data, disclosure, and accountability
Property firms handle exactly the data that makes the NIST emphasis on evaluation and use serious: identity documents, financial positions, and the addresses of empty homes. The controls are specific. Keep identity and financial documents out of general-purpose tools. Confirm with your vendor whether client data is retained or used for training. Log every automated message to a client so a complaint can be reconstructed. Disclose when a message was machine-generated, because clients discovering it themselves is a trust event you will not recover cheaply. And keep one named person accountable for anything a buyer or seller relies on. A companion on client trust is [Customer Journey Stories That Drove Real Change](/blog/customer-journey-stories-that-drove-real-change).
Section 6
What the research says
Small and mid-sized service businesses, the profile of most brokerages and property firms, are adopting faster than the headlines suggest. The U.S. Chamber of Commerce found 58% of small businesses now use generative AI, up from 40% a year earlier, and 82% of AI-using small businesses grew their workforce over the past year (U.S. Chamber of Commerce, 2025), useful evidence that automation and headcount growth are not opposites in client-facing industries. The drafting work that dominates real estate operations, listing descriptions, follow-up emails, viewing summaries, is exactly where controlled research shows gains: professionals using generative AI completed writing tasks 40% faster with 18% higher rated quality (Noy & Zhang, 2023). The risks are also well documented. Gartner predicts 60% of AI projects will be abandoned through 2026 if unsupported by AI-ready data (Gartner, 2025), and scattered listing data across MLS feeds, spreadsheets, and inboxes is the classic version of that problem. RAND's failure research adds that most AI projects fail because the problem was misdefined, not because the technology fell short (RAND, 2024), while only 39% of organizations report enterprise-level financial impact from AI (McKinsey, 2025). For a brokerage, the highest-evidence starting point is narrow: one transaction workflow, clean data, and a human reviewing anything a client sees.
Section 7
Metrics a principal should ask for
Not listings produced. Track time from enquiry to booked viewing, viewing to offer conversion, days from offer accepted to completion, fall-through rate, and administrator hours per transaction. Fall-through rate is the one that pays. Every deal that collapses late has already consumed the full cost of the transaction and returns nothing. If automated chasing pulls that number down even slightly, it has paid for itself several times over. Review quarterly as well as monthly, because property cycles are long enough that a single month tells you almost nothing.