How AI works in the price estimates
Understand the dynamic valuation model behind Property Valuation — comparable trades, individual adjustments, time indexing, location factors, seven condition levels and confidence.
The AI behind Property Valuation is not a single fixed number — it is a dynamic model that finds comparable trades in exactly the area where the home is located, adjusts each individual trade for differences, and calibrates the result across seven condition levels. This article explains what the model weighs, why the estimate looks the way it does, and how to read the confidence behind the number.
1The model starts with comparable trades
Once you have searched an address and calculated the valuation in Boligvurdering, the model pulls real, recorded trades in the surrounding area and uses them as its data foundation. It does not weight all trades equally — nearby, recent and more comparable trades count for more.
The model scales the number of trades up and down with how active the area is. In a dense urban market with many trades, a tight selection of the most comparable ones is used; in sparsely populated areas the radius expands (up to several kilometers), so there is always enough data behind the estimate.
2Each trade is adjusted individually
Instead of taking a raw average, the model adjusts each individual trade for the differences between it and your home. The adjustments are reflected in Detailed calculation for this property, and you can see the total adjustment per factor:
The factors cover, among others, time, size, floor, build year, noise, plot size (for houses) and distance. Coastal proximity is handled by weighting trades with a similar distance to the coast higher, so any waterfront premium is reflected automatically through the comparable trades.
The numbers in the card are an example. For your own home, shows the actual percentages — for instance how many dB of environmental noise the property is registered at, and how large the price effect is.
3Time indexing brings old trades up to today
A trade from two years ago does not reflect today's market. That is why the model time-adjusts each trade individually to the current price level for the home's postal code, based on price development over the most recent quarters. This means older trades can still be used — they are simply indexed up (or down) to present-day level before entering the calculation.
4Location adjustments: noise, roads and coast
The model accounts for nearby physical conditions that affect the price:
Noise — trades in noisy areas are normalized against a 55 dB threshold Roads — from a quiet residential street to a motorway, the property's location is adjusted Coast — trades with a similar distance to the coast are weighted higher, capturing the waterfront premium
These adjustments mean that two identical homes on the same street can receive different estimates if one sits on a busy road and the other on a quiet residential street.
5The result: seven condition levels
The model does not deliver a single number but a distribution across seven condition levels — from Poor condition, through Average condition and Very good, up to Luxury at the top. Each level is calibrated against real percentiles in the trades found, so you can place the home based on its actual condition:
In you also get a short text that places your home on par with comparable trades.
6Confidence: how reliable is the estimate?
Every estimate is accompanied by a Confidence score that reflects the data quality in the analysis area — number of trades, geographic spread, price variation and temporal coverage.
An estimate with many nearby, recent trades gets High confidence; an area with few or scattered trades is shown as Medium confidence or Low confidence. Use the score as a guide for how much weight to place on any single number.
An AI estimate is a basis for decisions, not a definitive answer. With low confidence you should review the comparable trades yourself and adjust the condition level to match the home's actual state.
7Help the model improve
Below each valuation you'll find Did we get it right? Here you can mark Yes, spot on or No, off target and enter what you believe is the correct price. Your feedback feeds into the ongoing calibration of the model.