Automated valuation models (AVMs) have been around for a while in the residential real estate industry. The first publicly-available AVM was released by Zillow in 2006. Interestingly, the goal behind creating this valuation algorithm was not to actually help people value their homes, but to create controversy. Zillow's founders later admitted that they thought showing people's estimated home values would create a stir and drive more traffic to their website. They were right. On the first day that “Zestimate” was released, the website saw over 1 million views. By the fifth day, views had reached 2 million. Since then, Zillow has grown to become the most popular residential real estate website in the United States.
Automated valuations haven't had the same effect on commercial real estate. Valuing commercial real estate is much more complicated than residential real estate. There are rent rolls, lease comparisons, construction costs, and other factors that aren't publicly disclosed. Plus, few people care about the value of their office building or their favorite shopping center as much as their neighbor's house. But that doesn't mean AVMs can't help commercial real estate. In fact, some of the world's largest valuation service providers already use AVMs.
“We've developed an AVM that's robust enough to serve even the largest clients with real estate assets around the world,” says Charles Fisher, director of value and risk analysis at JLL Risk Advisory. There has always been concern that commercial real estate valuation is too complex to be assessed well by algorithms. “AVM is accurate enough to help asset managers understand which properties pass their internal risk metrics and which don't,” Fisher says.
AVMs are used primarily as a way to quickly identify properties that may be undervalued or overvalued, rather than as a standalone method for commercial property valuations. AVMs can ingest vast amounts of information and provide valuation recommendations for hundreds or even thousands of properties in a very short period of time. This speed allows some companies to get ahead of their competitors by acquiring properties before others realize their potential or by selling properties before the market downturns. “Investment decisions will be made much faster thanks to AVMs,” says Fisher.
Like any technology, AVM will only get better over time. There will be plenty of new technologies emerging that help computers more accurately assess a property's condition and the market. As more historical data and contextual information is incorporated into the automated analysis, AVM will start to get closer to replicating the current appraisal process. “Eventually, computer vision technology will be used to analyze photos of properties, allowing AVM to understand the uniqueness of the location and the condition of the building itself.
Artificial intelligence and machine learning techniques also have the potential to improve commercial real estate valuations by finding patterns and insights in large data sets that human analysts may overlook. But for these AI models to be truly effective, they need access to vast amounts of highly structured data that encompasses all the granular factors that can affect a property's value – data at scale that the commercial real estate industry currently lacks. Some markets may not have access to as much data as others, and there can be a significant time lag between when significant new information emerges and when the data is incorporated into the models.
From the terms of a specific lease to a qualitative assessment of a building’s condition, there are plenty of nuanced details that a trained appraiser considers, but translating them comprehensively into a data format that a machine can process is extremely challenging. Finishes, amenities, views, and the overall “pride of ownership” a property exhibits can have a significant impact on its market value, but explicitly codifying those attributes into structured data is a major challenge. With limited detailed data available as input, AI valuation models risk being overconfident in their predictions or overlooking important nuances that a human expert would identify, creating potentially significant blind spots. To address this, JLL’s AVM comes with a confidence score to help managers understand the extent to which they need to re-check the work done by the algorithm.
“I think there will always be a need for humans to be involved in the evaluation process,” says Fisher. “We really believe in the general theory that AI will replace tasks but not roles. You'll still need people to oversee the evaluation role, but you'll be able to outsource a lot of that task to AI.” AI and machine learning can automate analysis at scale, but human expertise will still be crucial to validate results and fill in gaps that arise from limitations in data quality or implicit assumptions that models fail to capture.
Valuing commercial real estate is not as simple as estimating the price of residential property, and that won't change in the future. But that doesn't mean the AVM technology that has transformed the residential real estate landscape has no value in commercial real estate. Human valuations are more accurate than AVM valuations, but the gap is closing. As AVMs become more accurate and easier to use, smart commercial real estate investors will turn to them for a competitive advantage in acquisitions and to better understand the properties they already own.