Using AI for Commercial Real Estate Valuations
Table of contents
Browse through a collection of quotes from astronauts who view the earth from space and several common themes stand out. The first is why – as humans – we all haven’t done a better job of coming together and solving the world’s problems. The second, somehow related to the first, is the fragility of our planet and our very existence.
Then there’s China’s first astronaut, Yang Liwei, who remarked, in that endearing manner that only a Chinese person could, that “the scenery was very beautiful but I did not see the Great Wall.” Compared to Mr. Liwei’s vision, today’s satellites have become incredibly powerful. Imaging data is being used to count cars in store parking lots, to measure the depth of storage tanks that hold crude oil, and to map out small plots of land that are in various stages of farming rice. There seems to be an ever-increasing number of companies that are finding innovative uses for imaging data, and today we’re going to look at one that uses satellite imaging as just one data input for calculating commercial real estate valuations using artificial intelligence (AI).
Commercial Real Estate Valuations Using AI
Founded in 2014, Dutch startup GeoPhy took in their first disclosed funding round of $33 million just days ago to provide insights into the value of properties by combining a variety of different datasets – including satellite imagery – and then analyzing all the data with artificial intelligence algorithms. People have historically relied on the likes of real estate agents to help assign property values using comparables, and the problem with that approach is that all the estate agent cares about is closing that transaction so they can collect a commission. While it’s true that tools like Zillow are changing that, there are an increasing number of big data sets that can help provide a great deal more granularity when it comes to accurately assessing a property’s value.
While getting a property appraisal used to take weeks, from initial assignment to final report, GeoPhy eliminates that lag in information flow with instant results from their AVM (Automated Valuation Model) which is dynamically updated with new market transactions on a daily basis. After all, a valuation isn’t accurate unless someone is willing to pay it. The result is an average predicted value that is just 5.85% from the actual transaction price, an accuracy that is twice as good as traditional appraisals for commercial real estate. That level of accuracy means that their platform is now being used by some of the world’s biggest lenders and investors with a client list that includes names like Goldman Sachs and UBS.
Big Data for Property Valuations
The company argues that the challenge is no longer how we can collect the data – it’s all there for the taking – but how we can put all the data together to add value. For example, if you’re thinking about moving into a new office and you want to know how accessible it is for your employees, GeoPhy can help. The company provides an “accessibility score” per office that is calculated with no need for human input and takes into account variables like the number of public transport stations nearby, the intensity of the traffic, and the proximity to highways.
The real power is evident in the ability to aggregate that data at the company level. Every day, over 3,000 variables for the 300 listed real estate companies worldwide are collected, including their 200,000 underlying buildings, and real-time valuations are displayed on a dashboard. Think about how valuable that might be if you’re a private wealth manager in charge of a property portfolio for some neurotic wanker who calls you every time something negative airs on the BBC about London’s declining property values.
That’s some hard data you can come to the table with, innit. One Dutch merchant bank, Kempen, found that the number of global real estate companies that can be effectively tracked and monitored in detail has increased fivefold, from 60 to 300. The system can also take into account geopolitical events, like judging the impact of Brexit on the City:
If you’re at all interested in the methods used to calculate commercial property values, it’s a complicated process because of the unique challenges being faced when trying to value commercial properties. In the case of residential properties, it’s easier to make comparisons because all the properties have common attributes (size, number of rooms, proximity to schools, etc.) not to mention there are a lot more transactions you can use for comparables. The factors used to determine commercial property values differ significantly, and one example they give is crime.
Advanced Valuation Methods
It’s all about tradeoffs. For example, a commercial office building in the middle of Oakland California might be the most accessible when you look at your database of employee addresses. The problem is, it’s a place so barren that the only thing growing is the crime rate. Instead of Oakland, you could choose a nice commercial building in Walnut Creek where there is less crime but then everyone will complain about the trek. When you start doing tradeoffs like these with multiple variables, things can get complicated quickly. That’s why the company’s machine learning algorithms use a technique called (SHapley Additive exPlanations) or SHAP which they detail in a great article on Medium. The below diagram shows variables that have a meaningful impact on the valuation model:
Let’s take status for example. On the far left, you can see how the value becomes redder. (If you’re color blind, this simply means that for higher status areas, you pay more for less the further up you go. In other words, it’s not linear.) What this means is that high-status areas aren’t giving you the most bang for your buck. If you think about it, this makes sense. Wealthy people see the value of money differently from us serfs. In fact, the model actually ran into a brick wall trying to explain how astronomical some of London’s property prices are. In other words, the model understands property markets extremely well.
Conclusion
We previously talked about 9 Startups Using Artificial Intelligence in Real Estate, and now we can add to that list yet another use case for AI, commercial property valuations. The best part is that over time, the valuations will become even more accurate. That’s because unlike estate agents, machine learning gets more intelligent over time. It also won’t use manipulative language like calling a dilapidated building “charming.” Couple that constant increase of intelligence with new data sets being added to the platform and it’s easy to see why investors funded this very interesting Dutch AI startup.
Share
