How image analysis is changing property valuation in Sweden

A property's value is more than its square footage, floor level, and address. Every agent who has ever walked into a home and immediately felt it knows that. That problem has long been impossible to solve with traditional valuation models. Not because no one understood it, but because no one had a good way to measure it.

That is changing.

What data doesn't see — but images do

Classical automated valuation models, so-called AVMs, are built on transaction data. Sale price, size, location, tenure type, floor. That is a robust foundation — but it does not capture the newly renovated kitchen, the worn bathroom untouched since 1987, the ceiling height that makes the living room feel twice as large as it is, or the view over Riddarfjärden.

This is precisely where image analysis comes in.

By training computer vision models on property images, we can begin to extract attributes that previously lived only in an agent's gut feeling. Is the kitchen new or old? Are the surfaces in good condition? Is it a bright or dark home? Is there a balcony with a view, or does it face a courtyard wall?

Each of these factors affects the price. Now we can start measuring them.

What we do at Nezto

At Nezto, we have built a computer vision pipeline that analyses property images by room type — kitchen, bathroom, living room, bedroom, balcony — and extracts structured attributes from each image. The output feeds into our valuation model as a complementary signal alongside traditional transaction data.

This means the Nezto Vision Engine™ does not only know that the apartment is 72 sqm in Södermalm. It also knows that the kitchen is newly renovated, that the bathroom is of high standard, and that the living room has unusually good natural light. That information shows up in the valuation.

Why Sweden is an interesting market

The Swedish residential property market has characteristics that make image analysis particularly powerful here. The tenant-ownership market is dominated by a handful of large listing platforms, meaning image data is relatively accessible and structured. Prices are high and volatile, which means marginal quality differences carry significant economic weight. And agents are professional operators with high demands on supporting material — they want precision, not just an estimate.

That creates a clear demand for what we are building.

The challenges are real

We do not want to paint too simple a picture. Image analysis in a property context is technically complex. Images are taken with different cameras, in different lighting conditions, from different angles and for different purposes. A staging photograph is not the same as a documentation photograph. Models must be trained on relevant data and validated carefully to be reliable.

That is why we work closely with methodological expertise — including Professor Mats Wilhelmsson from KTH — to ensure that the signals we extract are genuinely statistically meaningful, and not merely visually impressive.

Credibility takes time to build. We are taking that time.

Looking ahead

Image analysis is not a replacement for traditional valuation data. It is a complement — a way to begin capturing the dimension of property value that has so far fallen outside the reach of models.

We believe this is one of the most important shifts in automated residential valuation over the coming years. And we are glad to be part of shaping what it looks like in Sweden.