Our property valuation model now runs in production for the Latvian market. Extending it to Spain is less about copying the model than about rebuilding what it learns from.
Verteo, our machine learning model for property valuation, is now in production in Latvia. It estimates the value of individual properties from market data and puts that estimate where lenders need it: into collateral assessment during mortgage underwriting, and into credit scoring as an input.
Running in production changes what a model has to be good at. A forecast in a notebook only has to be accurate on average. A valuation that sits inside an underwriting workflow is judged case by case — including in the places where the market is thin and comparable sales are few. That is where most of the engineering effort goes.
We are now extending Verteo to Spain. That is not a matter of copying the model across: the Spanish market is larger, more regional and structured differently, and so are its data sources. The work is in rebuilding the training data and the features for that market, then validating the model against it before it informs a single decision.
Because Verteo connects directly to our lending platform, every market it covers is also a market where a mortgage application can be scored against a current, model-driven valuation rather than a stale appraisal.