From data to deals: Machine learning’s impact on real estate

This November, Inman offers a deep dive into the world of proptech and the state of the startups that are building the future now. We’ll also debut a brand new coveted set of awards, Proptech All-Stars, celebrating the entrepreneurs, VCs, and visionaries in the field. It’s Proptech Month at Inman.

Whispered tales echo in the halls and speak of a crystal ball that, when peered into, can reveal glimpses of the unseen. Imagine, for a moment, having such an artifact on your desk. With a focused gaze, you see which neighborhood will shimmer with the most appreciation in the coming years.

Another glance, and the ball unveils which of your leads, like embers about to burst into flame, is finally ready to list. Let’s peer deeper, and perhaps you discern the exact trends that may influence a buyer’s choice, the ideal pricing strategy to generate a collection of offers on a century-old home, or the impending doom encroaching upon a home inspection. 

What if the ability to harness the power to predict wasn’t hidden in a mythical artifact or relegated to mere reverie? What if the answers and certainty a crystal ball could provide can instead be found in something more tangible and within reach? 

As we watch the technological evolution of artificial intelligence (AI) unfold, with every passing day, the idea of a modern-day crystal ball in real estate moves further out of the realm of fantasy and into the space of reality through the power of machine learning. 

So, what is machine learning? 

AI is a term that has been broadly used of late but is generally understood to be a machine’s ability to replicate human cognitive functions such as problem-solving, interacting with an environment, creativity, and perception. Machine learning is a subset of AI, in which systems continuously learn and improve through the analysis and consumption of data. This allows them to dynamically make decisions and/or predictions without the need for explicit programming on how to do so. 

With regard to how this translates into practical application, much of the technological underpinnings in our modern experience are driven by traditional algorithms. Algorithms are, in essence, sets of rules or procedures. These can be the manual steps we follow when baking a cake, for example, but are commonly referred to in the context of programming computers to complete certain tasks. 

While algorithms have become immensely sophisticated over time and serve as the foundation for most of our digital world, powering everything from complex computational models to the posts we see in our social media feeds, they have historically lacked the adaptive intelligence needed to respond to unanticipated variables or account for nuance. 

For example, imagine a basic self-driving car operating on the confines of a traditional algorithm. The algorithm will be programmed to say: Stop when the traffic light is red, go when it’s green, and slow down when it’s yellow.

If an unexpected situation arises — a truck barreling through the intersection when the light is green, for example  — if the system doesn’t have a rule to handle this scenario, the car may still drive through the intersection simply because the light was green. This is where technology has historically deviated from human-decision making processes.

In contrast, a self-driving car powered by machine learning wouldn’t operate solely on predefined rules. Instead, it would be given vast amounts of driving data, scenarios and outcomes to process. This allows it to virtually “experience” kids chasing balls into the street, cyclists making unexpected turns, broken stop lights, etc.

When the light is green and the aforementioned truck comes through the intersection, the machine learning-powered car, learning from past scenarios, can predict and react to this anomaly by slowing down or taking some type of evasive action, and can do so regardless of whether a programmer explicitly accounted for that scenario.

While the car operating on traditional algorithms strictly follows its coded instructions and lacks flexibility, the machine learning-equipped car constantly learns and adjusts its reactions based on countless situations it has been trained on, similar to a driver who adapts and anticipates based on years of experiences and continuous observation of his surroundings. 

Applying machine learning to real estate

Machine learning isn’t just a concept of the distant future. It’s a present-day reality and an ever-increasing component of tomorrow. There are numerous ways it’s already being incorporated into daily operations and strategic initiatives across various industries and has already made its way into various technological functions within real estate. Here are a couple of examples of how it’s being used today:

• Image Analysis: Machine learning enables a computer to teach itself about the context of visual data and interpret one image from another without explicit programming about that particular image. In real estate, this is being used to sift through listing photos and other visual data sets to identify features and attributes of a property, including things like finishes, outdoor features and even property condition. These findings can be applied to various use cases, including delivering personalized property suggestions for buyers.
• Automated Valuation Models (AVMs): The models used to estimate the value of a property continue to evolve rapidly from their early, more confined iterations. Rather than relying entirely on highly static rules (e.g.: add $25,000 in value for each additional bedroom) or limited sets of high-level parameters (e.g.: beds, baths, square footage), machine learning has enabled the use of much larger arrays of data, including neighborhood information and historical activity, to account for greater degrees of nuance. Image analysis is also driving comprehensiveness and accuracy forward in this area as well.

As the technology continues to evolve and expand, the capabilities of machine learning will allow for even greater depth and precision. With AVMs for example, we may see the integration of real-time neighborhood activity, upcoming infrastructural developments, or even sentiment analysis from residents of the area (referencing things such as social media, local news, online forums, etc.).

This will drive an even more comprehensive analysis of value at the property level, and potentially drive more accuracy around estimates of appreciation at the property and neighborhood level as well. 

With image analysis, we may see an evolution from identifying features to identifying wear and tear at a component level, such as using aerial photos of the property to measure the condition of the roof. Combining such findings with machine learning models that can make maintenance predictions (by analyzing replacement schedules, costs, etc. across national data sets), would not only inform the value of the property to some degree but also help real estate professionals and homeowners make informed decisions and effectively plan ahead. 

Machine learning has the potential to elevate virtual tours by tailoring them in real-time to a buyer’s preferences, perhaps spending more time in the kitchen area for a buyer whose searching behavior has indicated that to be a primary area of importance. Mortgage companies may use machine learning to better understand long-term financial behaviors of loan applicants, allowing for nuance in mortgage approvals and flexibility that incorporates an individual’s future financial trajectory.

The analysis of historical data on lead conversion, historical buyer/seller attributes, real-time lead behavior, market analysis, and sentiment analysis can all be joined together to deliver predictive analytics that help agents not only qualify leads but dynamically prioritize them and proactively engage them at strategic intervals.  

While that transformative power is already evident in today’s real estate landscape to some degree, we’re merely scratching the surface of its potential. The tools and applications we see now represent the early stages of what promises to be a profound evolution in the industry. Even what we anticipate to be a reality in the future represents only a fraction of what’s possible. 

That said, as groundbreaking as machine learning may be for the industry, no technology is infallible. In the same way weather forecasts, despite their advanced prediction models, occasionally miss the mark, machine learning, even in its most advanced state, will not always hit the bullseye. Real estate is and will remain a deeply human-centric industry. While machine learning can identify patterns and suggest probabilities, it is the agent who understands the emotions, motivations and stories behind each sale or purchase. 

While at times it may feel like peering into a crystal ball, especially as it reveals predictions and patterns that were previously hidden to the naked eye, machine learning will function more as an incredibly powerful assistant to the modern agent rather than a true crystal ball.

Machine learning can provide the data, precision, speed and scale to support and enhance what already exists in the agent’s intuition and the modern real estate agent’s ability to leverage these powerful tools will serve as a formidable ally in the quest to distinguish from the competition and deliver unparalleled service to clients.