I’m known in my consulting work as someone who pushes agencies toward technology adoption. I’ve helped dozens of offices modernise their systems and embrace PropTech tools. So it might surprise you to hear me say this:

When it comes to property valuations, I still trust good human appraisers over AI. Most of the time.

This isn’t nostalgia or technophobia. It’s based on watching how both approaches perform in the real world, across hundreds of transactions over the past three years.

The Case for AI Valuations

Let me start by acknowledging what automated valuation models do well, because they do plenty well.

Speed and consistency: An AI can process every comparable sale in a suburb in seconds, weighting each by relevance, recency, and similarity. A human appraiser does this intuitively but can only consider a fraction of the available data.

Removal of bias: Experienced agents develop patterns of thinking that can become biases. We might consistently over-value certain street addresses we’ve sold on before, or under-value property types outside our comfort zone. AI applies the same methodology everywhere.

Market tracking: When clearance rates shift suddenly—as they did in early 2024 when rates changed—AI systems adjust faster than human mental models. The algorithms don’t hold onto last month’s market sentiment when the data says it’s changed.

Objectivity for difficult conversations: When a vendor’s price expectation is unrealistic, having three independent AI valuations aligned below their target makes the conversation easier. It’s not just the agent’s opinion—it’s what the data says.

Where Human Judgment Wins

Despite all this, here’s why I still reach for experienced human appraisers in consequential situations.

Understanding Buyer Psychology

A beautifully renovated terrace in Paddington might have similar specs to a neighbour’s recently sold property—same bedrooms, bathrooms, land size. The AI values them similarly.

But the experienced agent knows that the renovated property’s kitchen opens perfectly to the north-facing garden, creating a light-filled entertaining space that buyers emotionally connect with. The sold comparable had a dark, enclosed kitchen that buyers tolerated rather than loved.

This isn’t captured in structured data. No API reports “emotional response to indoor-outdoor flow.” Yet it drives hundreds of thousands of dollars in price variation.

Local Dynamics Too Specific for Algorithms

Last month, an agent I consult with appraised a property in the Inner West. The AI models suggested $1.65-1.75 million based on recent comparable sales.

The agent knew something the algorithms didn’t: a specific buyer had been searching for exactly this property configuration for 18 months. Rare corner block, specific bedroom count, dual street access. They’d missed two similar properties and had explicitly told multiple agents they’d pay a premium.

The property sold for $1.92 million. The buyer was exactly who the agent expected.

No AI model can account for specific buyer demand that hasn’t yet manifested in sales data.

Properties That Break the Comparable Model

Automated valuations work by finding similar properties and extrapolating. But some properties have no true comparables.

Heritage properties with unusual configurations. Homes with significant development potential. Properties affected by zoning changes not yet reflected in sales data. Anything where the value depends on future possibility rather than present utility.

I’ve seen AI valuations miss by 30-40% on properties with development uplift potential. The algorithms valued the house as a house, missing that a sophisticated buyer was purchasing the land and approval potential.

Human appraisers who understand DA processes, strata subdivision requirements, and developer buyer behaviour catch what algorithms miss.

Condition and Presentation Nuance

Virtual inspection tools are improving, but AI still struggles to assess property condition accurately from photos.

A property showing beautifully in marketing might have issues visible only on physical inspection—poor quality finishes, wear patterns, deferred maintenance. Conversely, a property with mediocre photos might present far better in person.

Days on market correlate with presentation quality, and presentation quality affects value. This assessment remains firmly in human territory.

The Synthesis I Recommend

Here’s how I advise agents to combine both approaches:

Start with AI as baseline: Pull automated valuations from CoreLogic, PropTrack, and whatever other sources you have access to. Note where they agree and diverge.

Investigate divergences: When AI models disagree significantly, there’s usually a reason. Perhaps comparable selection differs, or one model has captured recent data the others haven’t. Understanding why helps your own analysis.

Layer human insight: What does the AI miss? Buyer pool specifics, pending developments, property-specific factors, presentation quality. These adjustments are where human expertise adds value.

Document your reasoning: When your final appraisal differs from AI suggestions, note why. This creates accountability and helps you learn from outcomes.

Compare against results: Track how your appraisals and AI valuations compare to eventual sale prices. Understand where each approach succeeds and fails in your specific market.

Where I Trust AI Over Humans

To be fair, there are situations where I’d take the AI over the agent.

When an agent has limited experience in a specific market segment, their intuition may be less reliable than algorithmic analysis of actual data.

When emotional factors might cloud agent judgment—a property they’ve driven past admiringly for years, or a vendor relationship that makes them want to deliver good news—the AI’s objectivity helps.

When markets are moving rapidly and recent data tells a different story than established mental models suggest.

And for routine valuations of standard properties in well-traded markets, AI accuracy is genuinely good. The human value-add diminishes when there’s nothing unusual about the property or market.

The Future Balance

I expect AI valuations to improve significantly over the next few years. Computer vision will get better at assessing condition from photos. Natural language processing will extract nuance from listing descriptions. Integration with planning data will catch development potential.

But I don’t expect AI to fully replace human appraisal judgment. The property market is fundamentally driven by human psychology—status, lifestyle aspiration, family dynamics, fear of missing out. Understanding these motivations requires human insight.

The best agents will be those who master both: using AI to handle data-intensive analysis while developing the human skills that machines can’t replicate.

That’s been the pattern with every technological change I’ve seen in 25 years. The tools change. The need for skilled professionals doesn’t go away—it evolves.

---

Linda Powers spent 25 years in Sydney real estate, including 15 years running her own agency. She now consults with agencies on balancing technology adoption with traditional real estate skills.