AI in real estate: listings, valuation and lead qualification
Where property tech benefits from language and vision models — and why automated valuation is harder than it looks.
Property data is unusually messy: the same building described five ways across four systems, with photographs of wildly varying quality and descriptions written to persuade rather than inform.
Listing quality is the quick win
Extracting structured attributes from free-text descriptions and photographs — bedrooms, condition, features, renovation state — turns an inconsistent catalogue into something searchable and comparable. It improves search, filtering and matching simultaneously.
Property AI: effort against reliability
Listing attribute extraction
- Reliability
- High
- Watch for
- Descriptions written to persuade, not describe
Lead qualification
- Reliability
- High
- Watch for
- Handoff losing context to the agent
Condition from photos
- Reliability
- Medium
- Watch for
- Staged, wide-angle marketing shots
Automated valuation
- Reliability
- Depends entirely on the market
- Watch for
- A number with no honest confidence interval
| Reliability | Watch for | |
|---|---|---|
| Listing attribute extraction | High | Descriptions written to persuade, not describe |
| Lead qualification | High | Handoff losing context to the agent |
| Condition from photos | Medium | Staged, wide-angle marketing shots |
| Automated valuation | Depends entirely on the market | A number with no honest confidence interval |
Automated valuation is a confidence problem
Producing a number is easy. Producing a number with an honest confidence interval, in a thin market, on a property with unusual features, is the actual problem. A valuation model that does not know when it is out of its depth is worse than none, because someone will act on it.
Vision for condition, carefully
Models can assess apparent condition and detect features from photographs, and photographs are marketing artefacts — staged, wide-angled, taken in favourable light. Treat visual assessment as a signal for triage, not a survey.
Lead qualification agents actually feel
- 1
Capture conversationally
Budget, timeline and requirements without a twenty-field form.
- 2
Answer from the listing
Grounded in real property data, with a clear handoff when it can't.
- 3
Book against real availability
Not a callback promise — an actual slot.
- 4
Route by readiness
With the reasoning visible, so the agent trusts the priority.
Lead qualification is where agents feel it
Agents drown in enquiries, most of which never transact. Conversational qualification that captures budget, timeline and requirements, then routes and prioritises, returns time immediately — and it is genuinely low-risk.
- Capture requirements conversationally rather than through a long form
- Answer property questions from the listing data, with a clear handoff
- Book viewings against real availability
- Score and route by readiness, with the reasoning visible to the agent