Most estate agencies now have an opinion about AI — some have adopted tools, some are watching, a smaller number have decided it is not relevant to them. What fewer have done is ask which problem they are actually trying to solve before choosing a solution. That gap is where most AI investment in estate agency goes wrong.
The category error
AI is not a single capability. It is a collection of techniques applied to different problems. Generating a property description is an AI task. Matching a buyer query to a listing is an AI task. Reading an inbound seller enquiry and assessing instruction likelihood is an AI task. These three tasks have almost nothing in common technically. The same label covers all of them.
When an agency adopts AI tools, they are usually solving a generation or retrieval problem: producing written content faster, or helping buyers find relevant listings. These are real improvements. They also have a ceiling. Better property descriptions do not change whether a seller will instruct. More accurate buyer matching does not reduce the number of valuation bookings that fail to convert.
The problem most agencies have not solved
The seller enquiry problem is not a generation or retrieval problem. It is a classification problem.
Every week, agencies receive enquiries from sellers at different stages of a decision. Some are ready to instruct. Some are six months away. Some are gathering information with no firm intention to move. These enquiries arrive through the same channels, use similar language, and typically enter the same response queue.
The agency that identifies the imminent sellers first, and focuses attention accordingly, converts at a higher rate. The research on this is consistent. Reapit data shows a seventeen-point valuation-to-instruction gap between average agencies and those in the top decile. Homeflow data shows that 39 percent of agents fail to respond to enquiries within twenty-four hours. The issue is not technology adoption. It is that the technology being adopted is not addressing the right problem.
What a useful question looks like
Before adopting any AI tool, an agency should ask what specific outcome it is trying to improve. If the answer is producing listing descriptions faster, generative AI is the right category. If the answer is improving buyer search experience, retrieval AI is the right category. If the answer is understanding which sellers are most likely to instruct and when, neither of those categories applies.
The seller enquiry problem has been unsolved for a long time. It is not unsolved because the technology does not exist. It is unsolved because the industry has been asking the wrong questions.
Agencies that start with the problem rather than the tool will find the answer is more specific, and more available, than they expected.