AI in real estate: where it actually changes the work
The agent doesn't disappear — the work around them does. Where AI is already paying in Australian real estate, what's overhyped, and what we'd push back on.
AI is not removing agents from real estate. It is removing the four hours of work around them each day. The biggest near-term wins are response speed and admin relief. A custom valuation model is probably not the right first investment for most agencies.
There's a lot of writing about AI "transforming" real estate. Most of it is wrong in the same way: it predicts the agent disappears.
That's not what we're seeing. The agent stays. The work around the agent changes.
What's already real
Three applications have moved from pilot to ordinary:
Valuation, with caveats. Automated valuation models now achieve mean absolute errors in the order of 10–11% in tested Australian metropolitan markets — close enough for guidance, not close enough for the final number on a unique property. A good agent's appraisal still wins on character homes, recent renovations, and anything off the comparable-sales grid. The right model is human-led, machine-supported.
Lead qualification. AI-driven scoring of online buyer behaviour — which listings they linger on, which floor plans they download — produces materially better signal than traditional digital marketing. Agencies using this well don't get more leads. They get the same leads ranked in the order worth calling.
The admin layer. Listing copy, contract drafting, follow-up sequencing, file management. This is the least glamorous category and probably the biggest near-term margin lever. The Australian agency average is something like 47 hours to first call-back on a portal enquiry. 78% of buyers go with the first agent who responds. The maths is obvious.
What's overhyped
A lot of vendors are selling "agentic" pricing engines that promise dynamic, real-time valuation recommendations. The technology exists. Whether your principal will trust a number generated by a model they can't interrogate, and put their reputation behind it at an appraisal — that's a different question.
We'd be cautious about replacing the appraisal conversation with a screen. The dollar value isn't the point of that meeting. The trust is.
Where the value is, in plain numbers
Two places, in our view.
The first is response speed. Sub-five-minute response to portal enquiries lifts engagement by something like 30%. For a mid-sized agency this is the difference between a normal quarter and a strong one, and it's mostly an automation question — not a hiring question.
The second is agent time. If an agent recovers four hours a week from listing copy, lead triage, and admin, that's two more inspections, two more appraisals, or four more genuine client conversations. At an average commission, the payback is not subtle.
Neither of these requires building anything. Both are off-the-shelf with thoughtful integration.
What we'd push back on
If a founder told us they wanted to build a custom AI valuation model, we'd say: possibly, but not yet. The base economics of a real estate business aren't constrained by the accuracy of an internal AVM. They're constrained by response time, appraisal-to-list conversion, and listings-per-agent. Fix those with workflow first. Build the model when you have data nobody else does.
That's the trade-off most agencies aren't being told: the right AI spend is rarely the most technically interesting AI spend.
