I’ve been watching automated valuation models (AVMs) slowly work their way into Sydney’s property market for about five years now. What started as a rough tool that banks used for internal risk checks has evolved into something much more significant. And it’s changing how properties get appraised in ways that affect buyers, sellers, and agents.

Let me walk you through what’s actually happening, because the reality is more nuanced than either the tech boosters or the sceptics want to admit.

What AVMs Actually Do

An automated valuation model takes publicly available data — recent sale prices, property characteristics, land size, location, zoning — and uses statistical models and machine learning to estimate a property’s current market value. The two biggest players in Australia are PropTrack (owned by REA Group) and CoreLogic.

These aren’t new. Banks have used AVMs for mortgage assessments for years, particularly for low-risk refinancing where sending a human valuer to the property isn’t cost-justified. What’s changed is the accuracy, the speed, and the range of applications.

Modern AVMs can pull in satellite imagery to estimate building footprints, renovation status, and even pool installations. They can analyse neighbourhood trends at a granular level — not just “Mosman is going up” but “this specific street in Mosman within 200 metres of the ferry wharf has appreciated 3% faster than the suburb average.”

Where They’re Accurate

For standard properties in data-rich suburbs, AVMs are surprisingly good. A three-bedroom house on a standard block in Dee Why, where there have been 15 comparable sales in the last six months? The AVM will probably get within 5-8% of the eventual sale price. That’s not perfect, but it’s useful.

They’re also excellent for tracking market movements in near-real-time. Rather than waiting for quarterly median price reports, AVMs can detect shifts within weeks. That matters when the market is moving quickly — which, in Sydney, is most of the time.

Where They Fail

The limitations are predictable but important.

Unique properties. A heritage-listed terrace in Paddington with a rooftop garden and harbour views? The AVM has very few comparable sales to work with. It’ll spit out a number, but that number could be 15-20% off in either direction.

Recent renovations. If you’ve just spent $300,000 renovating a kitchen and bathrooms, the AVM doesn’t know that unless the renovation shows up in satellite imagery or council records. It’s valuing the property based on what it looked like in the last available data snapshot.

Development potential. A 900-square-metre block that’s been rezoned for medium density is worth substantially more than the same block under single-dwelling zoning. AVMs are getting better at incorporating zoning data, but they don’t always capture the full development premium because there aren’t enough comparable sales of pre-development sites.

Strata nuance. Two apartments in the same building can have dramatically different values based on aspect, floor level, renovation status, and strata levy structure. AVMs struggle with this level of intra-building variation.

How This Affects Buyers and Sellers

For buyers, AVM estimates are a starting point, not a verdict. I tell my clients to check the PropTrack and CoreLogic estimates for any property they’re considering, but to treat those numbers as a broad range rather than a precise target. If the AVM says $1.8 million and the agent is quoting $1.95 million, that’s worth investigating — but it doesn’t necessarily mean the property is overpriced.

For sellers, there’s a growing challenge. Buyers walk into open homes with AVM estimates on their phones. If the automated number is lower than your asking price, you need to be prepared to explain why the property is worth more. That means highlighting features the AVM can’t see — renovation quality, natural light, neighbourhood character.

I’ve been working with Team400 on understanding how AI-driven data analysis applies to property markets, and one thing that’s become clear is that the best outcomes happen when human expertise and automated data work together. The AI handles volume and pattern recognition; the human handles context and judgment.

The Valuer’s Perspective

I’ve spoken with several registered valuers who are surprisingly positive about AVMs. Not because they want to be replaced — they don’t — but because AVMs handle the routine work and free up valuers for the complex assessments where human judgment actually matters.

The concern is that banks and lenders will over-rely on AVMs to cut costs, skipping human valuations even when the property warrants one. That creates risk — a bad AVM estimate on a $3 million property could lead to a $200,000 lending error.

What’s Coming Next

The next generation of AVMs will incorporate interior imagery from real estate listings, allowing the model to assess renovation quality and interior condition. Some companies are experimenting with natural language processing to extract value-relevant information from agent descriptions (“harbour glimpses” versus “panoramic harbour views”).

There’s also movement toward real-time comparable analysis that updates after every auction result, giving agents and buyers a constantly refreshed picture of where the market sits.

My Take

AVMs are a genuinely useful tool that makes the market more transparent. More data, available to more people, more quickly — that’s a good thing. But they don’t replace the knowledge that comes from standing in a property, understanding the street, talking to neighbours, and knowing the quirks of the local market.

The best approach is what I’d call informed scepticism. Use the automated data. Check it. Question it. Then layer your own judgment on top. That’s how you make good property decisions in a market as complex as Sydney’s.