Why the range matters more than the number

Type an address into any estimator and you'll get a single confident-looking figure. The honest version of that figure is a range: a model saying "if this property sold today, the price would most likely land between X and Y." The width of that range — not the midpoint — tells you how much the model actually knows.

Ranges are narrow when the evidence is dense: plenty of recent sales of similar homes nearby, a recent sale of the property itself, and attributes that sit close to the suburb's typical stock. They blow out when a property is unusual — a renovated terrace among unrenovated ones, the only five-bedder on the street, waterfront when nothing on the water has traded for two years.

That's why we publish the comparable sales behind every estimate. If the comparables genuinely resemble the property, weight the number. If they don't, treat the estimate as a starting hypothesis and get humans involved: a local agent for a selling appraisal, a licensed valuer where money is on the line.

A practical rule of thumb: high confidence plus a recent on-market sale of the property itself is about as good as automated estimates get. Low confidence on a one-of-a-kind home means the true price is decided at auction, not in a database.