The AI Ceiling in Real Estate: What Algorithms Still Cannot Price
Every dataset has an edge, a point where the numbers stop explaining the outcome. Wes McGregor, a REMAX Northern Realty principal with 25 years in the brand and nearly three decades working the Albany Creek market in Brisbane’s northern suburbs, has spent the last few years mapping exactly where that edge sits in residential real estate. His conclusion is not an argument against automation. It is a precise account of where automation’s returns diminish.
McGregor built Vaire, an AI platform that converts standard listing photos into cinematic, ready to post video content in under five minutes, cutting the time and cost of traditional video production dramatically. That is a clean automation win: a repeatable, structured task collapsed from days to minutes. But when McGregor gathered fellow REMAX directors in Queenstown and set them a deliberately provocative exercise, build a plan to fully replace the real estate agent with AI, the group stress tested the entire workflow and found a hard limit. Up to 95 percent of traditional agent tasks can be streamlined or automated. The remaining slice resists modelling entirely.
The first unmodellable layer is hyper local intelligence. Portals can calculate price per square metre, zoning classifications, and median yield with reasonable accuracy, and those figures genuinely matter for comparative analysis. But street level variance, which block backs onto a flood easement, which pocket has a school catchment quirk that never made it into any dataset, sits outside structured data entirely. McGregor describes 27 years of accumulated, unfiltered observation in a single suburb as a competitive asset that no portal training set currently captures. That is a useful reminder for anyone building or relying on property intelligence tools: hyper-local signal is often thin exactly where the data infrastructure is thinnest, and the gap does not close just because processing power increases.
The second layer is behavioural, not analytical: emotional volatility in negotiation. Pricing models can flag when a seller’s expectation diverges from comparable sales, but they cannot manage the seller once that divergence becomes personal. McGregor’s own description of the role is blunt.
Boiled down, our job as an agent is to save sellers from themselves.
What makes this framing useful beyond one agent’s workflow is the broader signal it carries about proptech adoption generally. McGregor calls the barrier cognitive dissonance rather than capability, most people default to fear when presented with a tool that is simultaneously useful and threatening, and only a minority engage with genuine curiosity. He expects the next 12 months to widen the gap between operators who build structured AI ecosystems, what he terms AIDAs, AI digital assistants layered across an entire practice, and those who avoid the shift altogether.
For anyone tracking the intelligence layer behind housing decisions, the lesson is not that AI plateaus at 95 percent automation by coincidence. It plateaus there because the remaining functions, hyper-local pattern recognition built from years of direct observation and real time emotional negotiation, are precisely the categories that resist structured data by nature. Efficient systems still need a human interpreting the noise at the edges. That distinction is where the next generation of property technology, and the professionals who deploy it well, will actually compete.
Source: Elite Agent, “Wes McGregor: AI, Market Shifts, & Agent Survival”
