We price homes with AI. Here is the moment we tell you to go and pay a human.

The number is the whole negotiation

Everything in a sale runs off the price you start with. Set it too high and you burn the first three weeks, which are the only weeks where buyer attention is free. Set it too low and you have donated the difference to whoever turns up first.

So the number matters more than the photos, the copy, the open times and the portal upgrades combined. Which is why it is worth being precise about how much confidence any given number deserves.

What our AI is genuinely good at

We run an automated valuation model over comparable sales data, and then we put AI over the top of it to read everything the model has no field for. Photographs. Floorplans. The condition of the kitchen. Room dimensions. What the owner has actually spent on the place since they bought it, and whether that spend shows.

That second layer is the useful part. An automated model knows your house sold for $780,000 in 2019 and that three houses nearby have sold since. It does not know you spent $95,000 on the bathrooms and the back deck in 2023. Feeding that in moves the number to where it should sit, rather than where the last transaction left it.

It is fast, it is consistent, and it has no reason to flatter you. An automated valuation does not want your listing. That alone puts it ahead of a lot of appraisals sellers get handed.

What it cannot see

I sat down last week with Craig Smith, director of Australian Valuers and a registered valuer working the South East Queensland market since 2008, and asked him where the models fall over. His answer was more generous than I expected, and then much blunter.

In a built-up area where there is a lot of available sales evidence and properties are turning over all the time, I have found them to be pretty good. The minute it is outside the box, they are just not reliable at all.

Craig Smith, Registered Valuer, Australian Valuers

Outside the box is the phrase to hang on to. Acreage. Coastal. Elevation. Anything where the value sits in something no database has a column for.

His example is the one I keep repeating to people.

You could be within the same street. One side of the street has ocean views and the other does not. A view could be $250,000 to $300,000 more in value.

Craig Smith, Registered Valuer, Australian Valuers

Two houses, same street, same era, same land size. A model reads them as twins. They are not twins, and the gap between them is bigger than most Australians’ deposits.

This is not an AI problem. It is a data problem, and it applies identically to every estimate you have ever seen on a portal. Craig’s view on buyers running properties through ChatGPT is that they are drinking from the same well.

It is probably drawing from exactly the same pool. So the limitations would simply be in line with those types of models.

Craig Smith, Registered Valuer, Australian Valuers

The confidence rating is the part nobody reads

Every automated estimate carries a confidence indicator. High, medium, low. Most people scroll straight past it to the dollar figure, which is the wrong half of the output.

The dollar figure is what the model thinks. The confidence rating is how much the model thinks you should care what it thinks.

A high confidence estimate in a suburb of near-identical homes with plenty of recent settled sales is a number you can plan a campaign around. A low confidence estimate on a five acre property with a shed, a dam and a partial renovation is the model telling you, politely, that it is out of its depth.

Our position is simple. When our confidence comes back low, we say so on the screen, and we would rather refer you to an independent valuer than let you launch a campaign on a soft number.

The bank does not care what you think it is worth

Here is the risk sellers underestimate, especially in a market that has stopped rising.

You can talk a buyer into a price. You cannot talk their lender into it. Craig put it in one line that I have quoted back to three vendors since.

You might be able to sell your property for more than it is worth, but if it is subject to finance, it is going to come unstuck at the end anyway. So why not establish that now?

Craig Smith, Registered Valuer, Australian Valuers

If a bank valuation comes in under the contract price, the buyer has to find the shortfall in cash or the deal dies, usually four or five weeks into a campaign, at which point you relist as a property that has already failed once. The market reads that. It prices it in.

A pre-sale valuation does not stop a bank valuer forming their own view. What it does is give you something to put in front of them if that view is wrong. Craig had it happen on a property he was buying himself, where an out of area valuer missed the fact the home had views and came in well under market. He sent them the existing valuation report. They changed the figure within seconds.

That is the actual mechanic. A seller with an opinion gets ignored. A seller with a registered valuer’s report and the comparable sales behind it gets the number reviewed.

I ran auctions for a living

I am an auctioneer by trade. I have stood in front of enough crowds to know that a room can invent a price out of nothing.

I have seen properties clear the quoted range by hundreds of thousands with no comparable sale anywhere behind the bidding. Two people who both decided this was the one, and neither of them checking the maths. Everyone congratulates the agent. Nobody checks whether the finance holds.

It cuts the other way too. On a quiet morning with three bidders and one of them serious, a good property can sell for less than it is worth simply because nobody in that particular crowd was prepared to go further that day.

What we actually recommend

The practical version, and it is deliberately not always buy the most expensive option.

Craig’s own test is a good one. If there are five houses on your street that are effectively identical to yours, they have all sold recently, and you are happy with those prices, list it and sell it. If not, you are guessing, and so is every free estimate on the internet, including ours.

Why we are telling you this

We charge a flat fee instead of a percentage commission. We do not earn more when your house sells for more, which means we have no reason to quote you a number designed to win a listing and then spend six weeks walking you back down to reality. Anyone paid on a percentage has that incentive whether they act on it or not.

The flip side of that independence is that we have to be honest about our own tools. AI valuation is very good at the common case and demonstrably weak at the unusual one. Saying so costs us nothing and tells you exactly how much weight to put on the number we give you.

If the number matters, and it always does, get it right before the campaign starts rather than after the finance clause fails.

We put the whole case to Craig in a separate interview, including what a pre-sale valuation costs and the one seller he says should not bother. That is here: Ask a valuer: is a pre-sale valuation actually worth it?.