Why online home estimates get it wrong

Automated home valuations are wrong often enough that everyone has a story about it. The interesting question is not whether they are wrong but why, because the reasons are structural rather than fixable, and knowing which one applies to your property tells you how much weight to put on the figure.

They cannot see the property

The single largest source of error. A model knows a house has four bedrooms and 1,850 square feet. It does not know the kitchen was last replaced in 1994, that there is damp in the back bedroom, or that the current owner is an unusually good gardener.

Two houses on the same street with identical recorded attributes can differ by thirty percent in condition, and no volume of data fixes that, because the information was never recorded anywhere.

Public records are wrong more often than you would think

Models are built on assessor records, deed data and listing history, all of which contain errors: wrong square footage, a bedroom count from before a conversion, an extension never permitted, a sale price that included furniture. Garbage in one field propagates through the whole estimate, and nothing in the output flags it.

Thin comparable evidence

Valuation works by comparison. Where similar homes sell frequently, models do well. Where they do not, the model is extrapolating, and it will do so without telling you.

This is why estimates are weakest exactly where people most want certainty: unusual properties, rural areas, converted buildings, anything at the top of a local market. If your house is the only one of its kind for a mile, treat any automated figure as a rough bracket.

Lag

Sale data arrives late. A US deed can take weeks to record; UK Land Registry data lags completion by weeks more, and completion itself lags agreement by a month or two. So an estimate is describing a market that existed some time ago. In a flat market this is irrelevant. At a turning point it is the whole story, and estimates are consistently late to both tops and bottoms.

The incentive problem

Worth saying plainly, because most sites will not. Many valuation tools exist to generate leads for agents and lenders. A homeowner who believes their house is worth more is more likely to enquire about selling. That creates a quiet, permanent pressure toward optimism.

There is no way to verify from the outside whether a given estimate is subject to it. What you can check is whether the site shows you how it reached the number. A figure with its arithmetic printed underneath can be argued with. A single number from an undisclosed model cannot, and you should discount it accordingly.

How to sanity-check any estimate in five minutes

• Ask what it knew. Did it use your actual purchase price, or guess from attributes? The former has a real basis.
• Look at the range, not the point. A tool quoting a range of plus or minus twenty percent is telling you something true. One quoting a single confident number is hiding the same uncertainty.
• Check it against real local sales. Sold prices are public in both the US and the UK. Find three genuinely similar properties that completed recently. If the estimate sits outside that spread, it is wrong, not the market.
• Check the date on the underlying data. If it is a quarter old and the market has moved, adjust your expectations before you adjust your plans.
• Ask whether your property is typical for its area. The more unusual it is, the less any area-based method can tell you.

What they are genuinely good for

Tracking direction over years. Establishing whether your equity is roughly where you assumed. Comparing areas. Deciding whether a conversation with an agent is worth having.

They are not good for pricing a listing, arguing with a lender, or settling anything legally. For those, pay for an appraisal. The fee is small next to the decision.