Machine Learning in Real Estate – a Game-Changing Opportunity

Over the past few years, machine learning has received extensive media coverage as it has shown to drastically improve efficiency in different industries, ranging from deep pharmaceutical research, FinTech, supply management, and since more recently, PropTech.

Real estate, being a traditionally fragmented and heterogenous sector, has seen a lag in technological innovation. This can largely be attested to the lack of structured datasets for real estate and has therefore resulted in a dispersed way in which information is presented.

Historically, real estate agents (also coined as brokerage in the US), have marketed and sold properties through traditional non-digitised marketing platforms such as newspapers. Through digitalisation, further marketing has boomed on websites and even on social media, adding to the wide range of sales channels being deployed. For each channel, brokers typically use their own style, using different forms of text, imagery or video, however the fragmented nature of this data has hindered the creation of structured datasets in the past.

Breakthrough discoveries within AI have now made it possible to create structured datasets. We are already at the stage where AI and machine learning technologies can outperform humans on a range of specialised tasks, such as information gathering, summarisation and image recognition. These systems show great potential for data aggregation but can also be deployed for applications in other disciplines.

Through this article, we highlight the potential applications of machine learning in the real estate sector, looking chiefly at: how machine learning can be deployed within real estate; how technological advances will impact the future of AI; and how technological advances will impact the real estate sector.

A brief history

To have a grounded understanding of machine learning, it would be unjust not to consider what has been achieved to date, and what has ultimately driven this success.

Neural networks (a name derived from the function of neurons in the brain), are interconnected cells that find patterns in data. Neural Networks initially received a lot of attention in the 1990s, but excitement died down when researchers realised that training these systems took too long, and parameters of large systems were difficult to estimate.

2012 sparked a change when researchers from the University of Toronto released Alexnet, a neural network for image classification which outperformed all traditional systems.

Their solution drastically reduced the training time required, generating huge amounts of interest in the foundations of AI, and its potential applications.

After the release of Alexnet, leading tech firms Google, Facebook, and Amazon actively competed on open dataset challenges to benchmark their state-of-the-art systems, which resulted in many improvements in the space. One of the most notable achievements was when AlphaGo, a program developed by Deepmind, beat the reigning champion of the game ‘Go’ in 2016.


With the recent technological advances in the areas of computer vision and language processing, it is now possible to utilise these advancements and apply them to real estate data.

Object detection systems, such as ‘Detectron’ developed by Facebook, are now capable of detecting objects from images. The real estate industry can leverage this to detect objects in residential and commercial properties, such as furniture, washing machines, refrigerators, and much more.

Image classifiers, like that of ‘Inception’ developed by Google, can be used to predict the architectural style or condition of the property, resulting in a multitude of applications that reach far beyond that of the real estate industry. In other areas, 2D floor plans of apartments can automatically be generated from videos submitted by landlords or agents, thereby eliminating the need for manual measurements and surveying, and greatly speeding up the marketing process.

Aside from images, significant progress is made in the area of information retrieval from text. There are machine learning systems that can directly answer questions based on text excerpts. Results on the “Stanford Question Answering Dataset” show that the top ranked systems have already surpassed human performance. The applications of these systems are vast and can essentially retrieve any information from the descriptions of listed properties, such as sales prices, conditions of the tenancy agreement, and state of the property.

The aggregated datasets allow people to create real-time visualisations of trends within the real estate market. Big data is also the baseline for automated valuation modelling (AVM), the process of estimating real estate prices from data. More importantly, machine learning has the ability to track consumer and investor sentiment, two important factors for predicting price movements.


“It is clear that a large number of machine learning applications are emerging in the real estate sector – resulting in a ‘big bang’ moment.”

Bart Melman

There are several drivers that have accelerated and will continue to accelerate its momentum.

The first driver is open source research. In computer science, almost all researchers have made their code and papers publicly available online. This transparency allows PropTech players to rapidly deploy state-of-the-art systems without replicating them from scratch, essentially giving companies a step up on innovation.

The second is the industrialisation of code, which allows engineers to utilise production-ready systems without requiring technical understanding and large investments. Even though this process has only just started, we see this as the real catalyst for greater innovation and applications of new technologies to this sector. The platform Allennlp exemplifies how Machine Learning as a Service (MLaaS) can bridge the gap between research and industry.

The third is hardware. Commentators believe performance of microchips and processors will continue to improve, with enhancements likely in 3D chip design, task-specialised chips and quantum photonic chips. These innovations will enhance performance overall and accelerate neural network training.

Market Impact

Over the past 5 years, the funding in European PropTech companies has expanded from $77 million to approximately $500 million, an increase of approximately 550%. The UK has taken a leading role in PropTech, with most emerging start-ups and the biggest investment volumes being derived from Europe.

The opportunities of cutting-edge machine learning technologies are widely recognised, where recent applications have moved far beyond automation.

A key use case of AI is seen in its application to the internet of things (IoT), where buildings are equipped with sensors to measure energy consumption, humidity, space optimisation, and other factors to understand the efficiency, quality and costs of the property in question.
Another key area we expect to have tremendous growth, are in AVMs, quantifying the sensitivity of property valuations to factors such as location, demographics and transport accessibility. Investors can utilise these insights to identify investment opportunities, perform risk assessments, and advanced portfolio management.

For the surveying profession, AI-powered innovations bring new kinds of efficiencies, which could of course impact jobs in this sector. A study by the Royal Institute of Chartered Surveyors indicates that 11 of the 42 key tasks undertaken by the surveying profession are at 100% risk in the coming decade from emerging technologies (including AI), and a further 29 tasks have at least 25% risk attached to them. According to the research, this requires a change in skillset and education of surveyors to work as data scientists or client managers to work with these new technologies.

The future

Advancements in machine learning have been made at rapid pace, and the opportunities for real estate will continue to grow over the coming years – and will likely be further accelerated by the COVID-19 pandemic.

“The aggregation of complex real estate data, previously deemed as too time consuming and difficult to extract, can now be directly translated into coherent and complete datasets.”

Bart Melman

This aggregation potential offers substantive and lucid analysis of underlying relationships that haven’t been explored to date.

At RE5Q, we believe that the innovative approaches to real estate have only just begun, where the true impact of machine learning has yet to be fully experienced. The use of open research has sparked numerous inventions in the fields of FinTech and PropTech, and it will continue to disrupt the industry over the coming years as a new era of data-driven real estate emerges.