How Close Is Our Property Estimate to the Real Price?
We tested on 182,517 homes under $800,000 across Australia. The average difference was $25,000.
Luke Metcalfe · Microburbs Research · March 2026The Problem
You search a property. You see three different estimates: one from the bank, one from a listing portal, one from a data company. They disagree by $80,000. None of them tell you how they got there or how often they are right.
We tested ours the hard way. We ran our model on 182,517 properties that had already sold, without telling it the price. Then we compared the estimate to what the property actually sold for.
Testing used five-fold geographic holdout: properties in the test set were spatially separated from those in the training set. This is more conservative than random sampling, which can allow nearby properties to leak across folds and inflate accuracy scores.
What We Found
| Price Band | Properties Tested | Within 10% | Within 5% | Avg Difference |
|---|---|---|---|---|
| Under $300K | 4,763 | ~78% | ~53% | $17,157 |
| $300K–$500K | 39,771 | ~83% | ~60% | $22,098 |
| $500K–$650K | 61,968 | ~88% | ~68% | $24,901 |
| $650K–$800K | 76,015 | ~90% | ~71% | $28,278 |
| All Under $800K | 182,517 | ~87% | ~67% | $25,495 |
Real Examples
Baldivis, Western Australia
1,006 properties tested · Median price $721,000 · Perth South West
Our best-performing suburb. About 96% of estimates were within 10% of the sale price. Average difference: just $19,761.
Why so accurate? Baldivis has uniform housing stock, regular sales, and consistent lot sizes. When every street looks similar, finding a good comparison is easy.
Frankston, Victoria
865 properties tested · Median price $730,000 · Melbourne South East
About 88% of estimates within 10%. Average difference: $31,939.
Frankston has a mix of older homes and new builds, which makes it slightly harder than a uniform suburb like Baldivis. But 865 recent sales means plenty of price references.
Caboolture, Queensland
820 properties tested · Median price $730,000 · Brisbane North
About 91% of estimates within 10%. Average difference: $28,512.
A classic investor suburb. Affordable entry point, strong rental demand, and consistent housing. Exactly the type of property our system values best.
When to Trust It, and When Not To
High confidence: 4+ comparables
About 9 in 10 estimates land within 10%. Use it for shortlisting, budgeting, and comparing properties against each other.
Medium confidence: 2–3 comparables
Useful for directional guidance. Could vary 5–15%. Good for filtering a long list, not final pricing.
Low confidence: 0–1 comparables
Treat as a rough indication only. This happens in rural areas, low-turnover streets, and very unusual properties. Consult local agents.
What This Means for You
If you are an investor buying under $800K: Our estimate is probably within $25,000 of what the property will sell for. Use it as a strong starting point for due diligence. Check the confidence level. If it says "high" with several comparable sales, you can rely on it. If it says "low", dig deeper before committing.
If you are a buyers agent: You can screen hundreds of properties quickly. Use the confidence level to prioritise which ones need a closer look and which ones you can price-check in seconds.
If you are refinancing or checking your home's value: Our estimate will tell you roughly where your property sits. Above $1.5M, accuracy drops to ~81% within 10%, with an average difference of $123,115. For anything above $1.5M, or if the estimate shows low confidence, get a formal valuation.
About the Research
We tested 12 different ways to improve accuracy. Only one made a meaningful difference: the prices of similar properties that recently sold nearby. Not the listing description. Not the suburb's crime rate. Not the distance to the train station. Just comparable sales. That is what drives accuracy, and that is what we focus on.
We tested whether adding listing descriptions, crime rates, school rankings, and transit proximity improved accuracy. None of them did — once you have comparable sales data, they add no meaningful signal. If nearby similar properties have sold recently, we can estimate accurately. If not, we cannot. That is the honest version of how automated valuations work.
Check a Property or Suburb
See our estimate for any Australian property — with confidence level shown.
Check a PropertyRead the Research