7,037 Melbourne properties tested head-to-head. 50,230 real sales across every Australian state. Full results published.
By Luke Metcalfe, Founder & Chief Data Scientist, 15+ years in property data analytics · 27 February 2026Every property portal shows you an "estimated value." You see it on Domain. You see it on realestate.com.au. You see it on ours. But here is a question the industry rarely answers with data: how often is that number right?
Not approximately right. Not "in the ballpark." Actually right. Within 5% of what the property sells for.
We decided to find out. We took our valuation model, MicroVal, and tested it head-to-head against Domain across 7,037 Melbourne properties where both AVMs had an estimate. Same addresses. Same sale prices from the land titles office. No cherry-picking.
One thing to note up front: we scraped Domain's valuations from sold property listings. We cannot confirm whether Domain's model updated its estimates after each sale was recorded. MicroVal's predictions are genuinely out-of-sample (the model never saw these properties during training). That difference matters when reading the numbers below.
The results surprised us. Not because we won (we expected to be competitive). But because of where we won and where we lost.
First, the headline numbers. For each property, we compared our prediction (MicroVal) and Domain's prediction against the actual sale price.
| Metric | MicroVal | Domain |
|---|---|---|
| Within 5% of sale price | 60.1% | 49.3% |
| Within 10% of sale price | 85.2% | 85.7% |
| Median dollar error | $37,626 | $49,799 |
| Closer on individual properties | 60.6% | 39.4% |
We beat Domain on precision. Our median error is $12,173 smaller per property. We were closer to the actual sale price on 60.6% of individual properties. At the tightest accuracy band (within 5%), we lead by 10.8 percentage points.
But Domain edges us at the within-10% mark by 0.5 percentage points. And they are better on expensive properties. More on that below.
Tables full of percentages are useful. But you want to know: does this work on actual houses? On streets you can drive down?
Here are real MicroVal predictions matched against what each property actually sold for.
35 Adelaide Lane
47 Contingent Street
5 Nagel Court
3 Lange Crescent
257 Grandis Boulevard
Across 50,230 matched sales, 30.3% of MicroVal predictions land within 5% of the final sale price. The median error across all properties is 8.8%.
MicroVal also provides a confidence range with every estimate. Here are examples where the sale price landed inside that range:
100 The Sanctuary Drive
12 Undara Road
18 Chitunga Road
In every case above, the actual sale price landed inside the confidence range. The point estimates were off by 1-3%. The range caught the real price.
We are not going to pretend we win everywhere. We do not.
Domain is more accurate above $1.2M. In the $1.5M+ bracket, Domain's within-10% rate is 93.6% compared to our 75.9%. That is a large gap.
Why? Microburbs is built for property investors. Most investors buy under $1M. Our model is trained, tuned, and validated for that price range. Domain is a broader property portal with more reason to optimise at the top end. Premium properties are also more individual. A $2M house in Warranwood might have a heated pool, a custom kitchen, and a view of the Dandenongs. Those are features our model does not yet capture.
In the $600k to $1M range, it is a different story. We win 68-74% of property-level comparisons in that band. Our median error is 3.5-3.7%. Domain's is 5.6-6.5%. That is the price range where most Australian investors buy. And that is where MicroVal is strongest.
In the $600k to $1M range, MicroVal beats Domain on 68-74% of individual properties. That is the price band where most Australian investors buy.
The testing is simple. We take properties that have actually sold. We check what MicroVal predicted before the sale. We compare the prediction to the recorded sale price from the land titles office.
No adjustments. No normalisations. No excuses. If the property sold for $860,000 and we predicted $848,702, we report an $11,298 error. If it sold for $1,434,896 and we predicted $1,145,533, we report a 20.2% miss.
We have done this across 50,230 real sales in every Australian state. And we have done a direct head-to-head against Domain across 7,037 Melbourne properties where we had both predictions for the same address.
Three things.
Do not trust a single number. Any valuation is an estimate. Look at the confidence range, not just the point estimate. If MicroVal says $800,000 with a range of $720K to $880K, the property could realistically sell anywhere in that band.
Compare valuations. Check MicroVal. Check Domain. Check CoreLogic if you have access. If all three agree the property is worth $950K and you are bidding $1.1M, that should give you pause. If they disagree, dig into why.
Be sceptical of round numbers. A model that says $848,702 has done the maths. A human valuer who says "about $850K" is rounding. Both might be close. But the model is showing you what it actually calculated, not what sounds plausible.
We are working on three things. First, extending MicroVal to properties that have not sold recently (right now, those fall back to a rougher estimate). Second, closing the accuracy gap on premium properties above $1.2M. Third, building a forward-looking benchmark where we make predictions before the property transacts, then check ourselves after.
We will keep publishing the results. If we get worse, you will know. If we get better, you will know that too.
The property industry does not lack confidence. It lacks accountability. We are trying to fix that.
You do not need to take our word for it. Every Microburbs Property Report includes a MicroVal valuation with a confidence range. Search any Australian address. You get the estimate, the confidence band, suburb analytics, comparable sales, and growth trends. All in one report.
Every Microburbs Property Report includes a MicroVal estimate with confidence range. Search any Australian address to see it in action.
