Last month, I ran an experiment that’s been brewing in my mind for two years. I took 20 properties across Sydney—from Bondi apartments to Parramatta townhouses—and got valuations from three AI platforms and three human appraisers I’ve worked with for decades.

The results weren’t what I expected.

The Setup

I selected properties deliberately to test edge cases:

• 5 standard 3-bedroom houses in established suburbs (Epping, Lane Cove, Cronulla)
• 5 apartments in high-density areas (Sydney CBD fringe, Parramatta)
• 5 properties with unusual features (heritage overlays, dual occupancy potential)
• 5 recent sales where I knew the actual settlement price

The AI tools I tested included CoreLogic’s automated valuation model, PropTrack’s estimates, and a newer entrant that uses computer vision to assess property condition from photos.

The human appraisers? Three agents with 15+ years experience each, operating in different parts of Sydney.

Where AI Got It Right

On standard suburban homes—the bread and butter of most portfolios—AI performed remarkably well. For a 3-bedroom weatherboard in Epping that sold for $1.82 million, the AI estimates ranged from $1.75m to $1.88m. That’s within 4% accuracy.

The machines excel when there’s abundant comparable sales data. Areas like the Hills District, where you have streets of similar homes selling regularly, are perfect for algorithmic analysis. Days on market for these properties tends to be predictable too.

CoreLogic’s system in particular impressed me with its ability to factor in recent infrastructure changes. It had already adjusted estimates near the new Metro stations before some agents had updated their mental models.

Where Humans Still Win

The heritage-overlay properties? The AI tools fell apart.

A Victorian terrace in Paddington with a development application lodged came back with wildly different AI estimates—a $600,000 spread between the highest and lowest. Meanwhile, all three human appraisers landed within $150,000 of each other, and within $100,000 of the eventual sale price.

Why? Because experienced agents understand buyer psychology. They know that a DA-approved property in the Eastern Suburbs attracts a specific buyer pool willing to pay a premium. They understand the strata implications of properties with shared walls and complex title arrangements.

One property in Surry Hills had significant vendor paid advertising in its recent campaign—over $40,000 in VPA that signalled serious seller intent. The AI had no way to factor this context into its assessment.

The Hybrid Approach That Actually Works

Here’s what I’ve started recommending to the agents I consult with: use AI as your starting point, not your conclusion.

The smart workflow looks like this:

1. Pull automated valuations from at least two sources
2. Identify where they agree and where they diverge
3. Investigate the divergences—they usually point to property-specific factors the algorithm can’t see
4. Walk the property with fresh eyes, knowing what the data suggests
5. Adjust based on your local knowledge and current buyer sentiment

This approach has cut my appraisal preparation time by about 30% while improving accuracy. The AI handles the comparable sales analysis that used to take hours of manual research.

The Clearance Rate Connection

There’s another angle here worth mentioning. Sydney’s clearance rates have been volatile this year, swinging between 55% and 72% depending on the week. AI valuations struggle to adjust quickly to these shifts in market sentiment.

When clearance rates dropped sharply in May, the human appraisers I work with immediately adjusted their guidance. The AI tools took 3-4 weeks to catch up, still suggesting prices that weren’t achievable in a cooling market.

This lag matters enormously for vendors making listing decisions. An overinflated AI valuation in a softening market leads to extended days on market, price reductions, and frustrated sellers.

My Verdict After 25 Years

I’ve seen technology transform this industry multiple times—from fax machines to the internet to smartphone apps. AI valuation is real, it’s improving, and it’s not going away.

But here’s what the PropTech evangelists often miss: real estate transactions are fundamentally human decisions. Buyers don’t purchase comparable sales data—they purchase homes where they imagine their lives unfolding.

The best agents I know are embracing AI tools while sharpening their uniquely human skills: reading buyer body language, understanding vendor motivations, negotiating outcomes that data can’t predict.

AI will keep getting better. But I’m not worried about robots replacing good agents. I’m worried about agents who don’t adapt becoming irrelevant.

The future belongs to those who can interpret what the machines tell us and add the human insight that closes deals. That’s not a contradiction—it’s the path forward.

---

Linda Powers has spent 25 years in Sydney real estate, including 15 years running her own agency in the Eastern Suburbs. She now consults with agencies on PropTech adoption and digital transformation.