When algorithms price the street

Automated valuation models have become fluent. They are not yet trustworthy at the border.

An algorithm trained on a national land registry can be unnervingly good at pricing a terraced house in a city it has seen ten thousand times. The same model becomes decorative the moment the comparable set crosses a language, a building code or a tourism overlay. A penthouse in Torrevieja is not a penthouse in Aarhus, even if both have three bedrooms and a terrace.

Satellite imagery, clickstream and listing velocity now dress these models in the language of certainty. That is useful for screening. It is dangerous as a substitute for a site visit, a lawyer and a person who knows which urbanisation floods in October. Cross-border buyers, especially those acquiring from Central and Northern Europe into Spain, are particularly exposed: they arrive with a number produced in another market’s logic.

The professional response is not to reject the machine. It is to put it in its place. Technology should shorten the distance between a Prague desk and a Costa Blanca plot — photographs, floor plans, Matterport, document packs, a common reference. It should not pretend that a yield calculated from portal averages is a substitute for the community fees, tourist-rental licence and orientation that decide whether the asset works.

REALNYX treats software as infrastructure for a human network, not as an oracle. The catalogue exists so that an office in one country can present an asset in another without losing authorship. The valuation remains a judgement. The platform’s job is to make that judgement better informed, not to replace it with a score.

The firms that will be punished in the next downturn are those that sold algorithmic confidence as if it were title insurance. The street is still priced by people who have walked it.

The street is still priced by people who have walked it.