Machine learning in real estate: an innovation catalyst

Machine learning in real estate: an innovation catalyst

August 26, 2021

Andrea Di Stefano

Technology Research Analyst

In 2018, a New York-based startup founded by four Israeli entrepreneurs called Skyline AI acquired two multifamily residential complexes in Philadelphia for $26 million. You might think that, so far, there’s nothing new under the sun, and you would be right. Except this deal was driven by artificial intelligence and, specifically, a machine learning algorithm capable of spotting real estate opportunities by assessing a property's market value and its possible future price.

This evaluation process was carried out taking into account a multitude of factors, including the asset's features, the local real estate scenario, the access to facilities, and many others.

However, the analytical capabilities of this machine learning system turned out to be much deeper than one might have expected, investigating even the most subtle nuances. For example, the algorithm recognized potential signs of mismanagement, "betting" on the fact that the owners would be willing to get rid of their property at a price lower than the actual value, consequently suggesting a lower bid. And it worked, as the company settled the deal for 12% less than previously expected.

Such a remarkable performance may suggest that the real estate industry should be increasingly relying on machine learning consultancy. Let's see how things stand at the moment.

Spicing up a conservative industry

As we already pointed out in our previous overview of RPA in real estate, this industry has traditionally approached new technologies with both curiosity and caution, but with particular emphasis, to be honest, on the latter.

This doesn't necessarily mean that you should think of real estate agents as cavepeople hopping around the black monolith in the opening scene of 2001: A Space Odyssey. However, such an ambiguous (not to say complicated) relationship with modernity was particularly evident in past years, so much so that the Morgan Stanley Digitalization Index ranked real estate as the second least digitized sector in the world.

It is difficult to find an all-encompassing reason for this conservatism, but we may speculate that an important role is played by the "human factor" and specifically by the immense variability of purely subjective parameters in the choice of something intimate and personal like a home. As much as we try to digitize part of this selection process, for example through the use of online platforms, customers will still want to evaluate their potential purchases in person and interface with a trusted real estate agent who acts as a guarantee between buyers and sellers.

Regarding the importance of personal relationships in this sector, the 2020 National Association of REALTORS® Profile of Home Buyers and Sellers showed that 89% of American sellers have called on a real estate agent when selling their home, and 41% of them selected their agent based on the recommendation of friends and family.

The granite traditionalism described above, however, seems set to falter, as the real estate industry has slowly begun to embrace some of the major trends of digitization.

A new technological impetus?

In recent years, the most far-sighted players in the real estate market have realized that artificial intelligence, machine learning, and other cutting-edge technologies can easily coexist with, or even enhance, the most human aspects of this sector, as well as leading to significant profits. Such ambitious aspirations have also prompted these firms to invest more and more in machine learning consulting.

According to statistics from CB Insights, indeed, tech companies that focused on the real estate sector have benefited from a skyrocketing funding growth between 2015 and 2019, with a slight decline in 2020 likely due to the recent pandemic outbreak.

Funding to real estate tech companies, 2015-2020

Other relevant insights in this regard come from Altus Group's CRE Innovation Report 2019, which highlighted how the adoption of AI and machine learning in commercial real estate portals was on the rise, with 48% of the respondents already using them in a significant or limited way.

Specifically, 19% of the companies surveyed deployed such technologies for analytical purposes, while 16% of them implemented artificial intelligence and machine learning to improve their accounting and property management processes. The idea behind this choice of AI use cases was to target those business functions that would potentially benefit most from the implementation of such tools.

Real estate business functions augmented with AI and machine learning

Since we've brought up use cases, let's briefly recap those where artificial intelligence, and even more so machine learning, has proven to be a real game-changer for the real estate industry.

Ready to enhance your business with machine learning?

Our experts will guide you

Contact us

‘The Price is Right’

When it comes to assessing the value of a real estate asset, we simply cannot rely on luck like in the famous TV game show hosted by Drew Carey, as the success of a good deal greatly depends on such estimates. This evaluation process has traditionally been based on a mix of intuition, professional experience, and analytical techniques, which took into account a relatively wide range of standard parameters. For example, real estate firms could compare a property with assets having similar attributes or located in the same area to get a general idea of the potential price.

However, the actual variables involved are so numerous that many of them easily escape this type of analysis. Luckily, the latest advances in machine learning and the growing availability of big data seem capable of offering an excellent solution to this problem. After all, machine learning systems and data match just as well as a good white wine pairs perfectly with fish. What does it mean in less gastronomic and more real estate-related terms?

Well, it implies that the increasing amount and variability of information regarding the real estate market can be processed by machine learning systems, whose algorithms will rummage through data looking for correlations or recurring patterns and build mathematical models representing those relationships. Once trained, these models will be able to process additional data, further refining their analytical skills and offering insight into current phenomena or predictions about future trends.

Oversimplifying through a pretty common example, imagine feeding our machine learning system with a large dataset relating to the square footage, number of rooms, and price of previously sold real estate properties. At this point, a linear regression algorithm will start processing this ensemble of information trying to find its underlying logic, which is likely to be what we may summarize as "bigger house, higher price."

Example of linear regression modeling for house price predictions

Later, we can use the model generated by the algorithm to establish the price of another property simply by providing it with the square footage and the number of rooms that characterize the asset.

If you're wondering what the deal is, as you could have gotten there on your own without the help of a machine, let me add that this process can be transposed on a much larger scale, embracing an immensity of additional variables. In this regard, it's worth opening a brief parenthesis on the role of non-traditional data.

Aiming at an overall vision

Would you buy a huge and relatively cheap house but, on the other hand, in poor condition and far from any essential facilities? If the answer is no, it's because, alongside pretty evident parameters such as those mentioned in the previous example, there are a multitude of not-so-obvious variables that still influence the attractiveness of a real estate asset.

In order to overcome this issue, real estate firms dealing with artificial intelligence and machine learning have started to consider a broader ensemble of key features defined as non-traditional data. This may represent, according to McKinsey, a source of information to fuel almost 60% of our potential predictive power.

Proportion of predictive power from traditional and nontraditional features

The list of non-traditional features which may have a relevant impact on the value of a real estate asset includes a huge amount of geographic, esthetic, and socioeconomic parameters, such as:

  • Properties' physical appearance
  • Transport network
  • Employment opportunities
  • Proximity to and quality of local services
  • Crime rate in the neighborhood
  • Parking availability
  • Greenery in the surroundings
Impact of non-traditional metrics on rental performance

How to get this data?

As you may expect, manually collecting such a broad spectrum of data from disparate sources would be a challenging task. Fortunately, artificial intelligence and machine learning can do most of the job for us, offering cognitive technologies such as computer vision and natural language processing (NLP) that already proved to be exceptional tools for automated data collection.

For example, photo or video object recognition technologies based on computer vision can be used to obtain meaningful insights into the quality of a property's neighborhood from visual data sources such as satellite imagery. Have you ever taken a virtual tour of a certain urban area through Google Maps' 3D visualization feature to choose a hotel to stay in? Well, imagine to perform this procedure several times faster and on a massive scale. That's what we're talking about.

Regarding NLP, on the other hand, it is possible to deploy this technology to gather detailed information on the characteristics of a property and the surrounding environment, including:

  • Sociodemographic data from government reports and statistical documentation.
  • Users' opinions (commonly referred to as sentiment) on nearby services and recreational activities from posts on AI-based social media, online articles, and review sites.
  • The attributes of other assets for sale in the neighborhood from the announcements published by competitors on their websites.

Once again, harnessing the knowledge hidden in this kind of data is basically a matter of pattern recognition, specifically of teaching machines how to identify recurring visual and semantic patterns.

For example, a machine learning system provided with textual descriptions and images gleaned from thousands of ads might understand that some particularly glitzy elements in a picture are typically associated with large luxury properties, while certain adjectives (something like "cozy", "lovely", “opportunity”) are commonly used to refer to much more affordable apartments.

Example of NLP in real estate advertisement analysis

The outcomes of this innovative approach, already embraced by various real estate valuation platforms such as Zillow and HouseCanary, have been extraordinarily promising. According to estimates reported on Forbes, machine learning-powered Automated Valuation Models (AVM) are able to assess the value of an asset with an absolute error inferior to 4% for homes and under 6% for commercial properties: a way more superior performance compared to that achievable with traditional methods.

Partner with Itransition to bring your machine learning project to life

Machine learning consulting

Full-spectrum ML services

Our machine learning service scope encompasses building, training and deploying ML models and recommendation engines for enterprise and customer analytics.

What about marketing and management?

Forecasting the value of a real estate property through machine learning superpowers without being able to effectively connect buyers and sellers is like purchasing a fridge in the North Pole: surprising, even fancy, but not that useful. This is when marketers come into play, and machine learning can lend a hand to them too.

Also in this aspect of the real estate industry, the mechanism is conceptually very similar to that described for pricing, and is commonly used when leveraging machine learning in ecommerce to power the well-known machine learning-based recommendation systems. This time, however, our goal is to use algorithms to segment potential customers based on their tastes, attitudes, and needs. The segmentation process is carried out through:

  • Processing user data related, for example, to their activity on ecommerce platforms or their Google search history.
  • Identifying recurring behavioral and purchase patterns in this information.
  • Classifying users into different archetypes to target them with real estate ads that may actually grab their attention.

If you allow me another (over)simplification, this implies that a real estate firm looking for a buyer for a beautiful mansion in Tuscany might decide to target wealthy managers who travel to Italy each summer for tourism, follow several YouTube channels on the Italian cuisine, and usually buy Italian-made products on Amazon.

Assisting managers with algorithms

Therefore, let's conclude our overview with some interesting applications of machine learning in property management.

In this regard, one of the advantages of using algorithms is the possibility to optimize energy consumption, achieving both savings in the assets' management costs and a reduction in their environmental footprint. A machine learning system trained with usual consumption data will easily recognize a shift from ideal conditions, such as a suspicious spike in energy use, and will alert management to quickly solve the problem. For example, the real estate investor KBS saved $47,000 in electricity costs at Premier Office Towers in Emeryville, California, thanks to machine learning-powered analytics.

Another scenario involving anomaly detection with machine learning concerns the deployment of this technology for predictive maintenance. This involves the rigorous condition monitoring of some key components of a property, including the electrical system, to detect any deviations from their standard performance and therefore predict potential failures before they occur.

Finally, machine learning can be implemented in real estate management to assess which asset improvements are worth investing in to significantly increase the value of a property. This idea has been the foundation of predictive analytics real estate platforms such as Enodo Score.

A fine blend of AI and human touch

By unlocking a full spectrum of exciting possibilities in terms of augmented data analytics, pricing, marketing, and management, artificial intelligence and machine learning have proved to be among the major catalysts driving a pretty conservative sector like real estate towards new technological horizons. PwC's Emerging Trends in Real Estate 2021 report, for example, ranked AI third in the list of the top tech trends disrupting real estate.

A fine blend of AI and human touch By unlocking a full spectrum of exciting possibilities in terms of data analytics, pricing, marketing, and management, artificial intelligence and machine learning have proved to be among the major catalysts driving a pretty conservative sector like real estate towards new technological horizons. PwC's Emerging Trends in Real Estate 2021 report, for example, ranked AI third in the list of the top tech trends disrupting real estate.

However, such a radical transformation may raise doubts and uncertainties, as such technologies are still far from perfect. In this regard, it is worth mentioning the case of Listing AI, an artificial intelligence-powered solution capable of autonomously writing descriptions of real estate properties for sales. While this supervised machine learning tool may be a valuable aid in speeding up the listing of huge numbers of assets, its computer-generated texts often ended up sounding rather contrived and therefore required engaging administration.

In other words, there’s still some way to go before achieving the delicate balance that will allow human experts and machines to work in full harmony. However, an important step has already been taken: setting in motion a real estate sector which, at least in technological terms, traditionally tended to be rather unmovable... like the assets it deals with.