How Machine Learning and Computer Vision are Improving Real Estate Valuations

Right now and for the foreseeable future, real estate is likely to go through something of a slump. As the world economy slows to a crawl and potentially dips into a recession, as a consequence of Coronavirus (COVID-19), real estate won’t be moving at the same pace.

However, real estate has weathered many a storm in the past. Many recessions have come and gone, with the sector recovering every single time. People are going to be busy moving home, and buying and selling once this is over (and in some cases, during this global pandemic).

Consequently, we still need to think about challenges in the sector and how we fix them moving forward. One of those challenges is the valuation of real estate.

Why real estate valuations are challenging

When it comes to buying and selling property, the valuation is key. It determines what a seller gets, and what a buyer needs to pay, and therefore who can afford a property and who can’t. But beyond the key players in that negotiation, the valuation impacts how much risk a lender is underwriting. Plus, the risk is equal to that for the insurance company.

Real estate valuations have a wide ranging impact across the economy. Automated Valuation Models (AVMs) are usually used to make decisions on the amount of finance being issued. It therefore directly influences loss mitigation and credit risk management.

Despite the importance attached to the valuation of real estate, industry experts know this figure isn’t always accurate. In many cases, the amount a property is valued by could be inaccurate as much as 15%. Hardly a small amount. Say a property is worth $100,000, it could be over or under-valued as much as $15,000. Now double or triple that for larger apartments and homes. All of a sudden, that 15% based on traditional AVMs has a serious impact on property affordability, and the underwritten risk being assumed by lenders and insurance companies.

Is there a solution to this valuation problem?

Yes, there is, and what’s more, it’s already being put into practice in the real estate sector.

Innovative and forward-thinking companies are already using machine learning (ML), and Computer Vision technology to improve the valuations of properties. Zillow is one such company that is making use of these solutions. Zillow has incorporated image analysis, which is a version of Computer Vision technology, alongside ML, to provide more accurate valuations within their Zestimate tool.

Zestimate is a way of using structured data, such as the square footage, number of bedrooms, number of bathrooms and location to provide a valuation. All of that information is compared to the price of other homes that have sold in the area, which is also structured and publicly available. Accounting for variations based on interiors and other changes homeowners make, this is the main way the sector works out valuations.

As we’ve already noted, too often this can lead to valuations that aren’t accurate, nor tailored enough for individual properties. Hence the problem of many being over or under-valued. Thanks to the reliance on AVMs, realtor judgments aren’t often enough to give an accurate price.

In 2016, Zillow started working on a “neural network” that could incorporate unstructured data to improve the accuracy of valuations. Over time, images started to be fed into this evolving neural network, also known as a machine learning (ML) algorithm. How homes are designed on the inside can make a noticeable difference.

For example, a property with granite kitchen countertops and stainless steel built-in appliances should fetch a higher price than one with a similar square footage and a Formica kitchen. Details matter, especially in real estate, which is why Computer Vision and ML is the most effective way to get the kind of valuation that realtors often overlook.

It isn’t just surfaces and stainless steel. Imagine if you’ve put a lot of money into home over the years. Decorating every room, ensuring the windows, doors, and other fixtures and fittings are completed to the highest standard. Another house down the street has stayed the same for over 10 years. Imagine then a realtor gives them the same value, simply because of the location, number of rooms and square footage. Wouldn’t you be rightly annoyed and upset?

Of course you would. Accurate valuations reward effort and investments made over the years. Hence the usefulness of adopting computer vision and machine learning for producing more accurate and fairer valuations.

How can real estate companies benefit from this?

Accurate valuations are a selling point and USP for sellers. Knowing that one real estate company is known for getting a true and worthwhile price is something that would attract sellers with more valuable properties. Not only is that good for business and brand reputation, but commission rates are going to be higher, thereby generating more revenue for doing the same work.

Don’t worry if you aren’t equipped with the tech capabilities of Zillow. Working with the right tech partner, neural networks and ML applications can be custom developed to give you a real competitive advantage.

With a wide range of computer vision, AI and ML-based solutions on the market, there are ways to produce more accurate valuations without hiring an internal team of developers and data scientists. We can create a customized solution that uses image analysis (computer vision) to determine the valuation of properties more accurately than are currently possible.