Nearly 200 years ago, one of the great scientists of his time, Charles Lyell, published, “The Principles of Geology.” In this influential work, Lyell extended the notion that changes in the geological record were gradual over time, giving credence to the notion of “uniformitarianism.”
Uniformitarianism was cast against the notion of “catastrophism,” which held that huge changes occurred as the result of massive stimulus and were few and far between. Lyellian gradualism was canonical, even in evolutionary biology, before an intervention in the early 1970s by Niles Eldredge and Stephen Jay Gould.
Eldredge and Gould posited that the geological record did not support gradualism in evolution and, instead, suggested what they called “punctuated equilibrium” in which massive changes happened over short periods of time; in between things were static. Gradualism, they argued, was not supported by the geological record.
Analogies are never perfect, but the debate over punctuated equilibrium versus gradualism reminds us of the world of technology and more specifically, of the effects advanced technology will have on the real estate industry. One of these epoch-making technological innovations is Computer Vision. Put simply, Computer Vision will create massive change over a short period in the real estate industry, similar to the punctuated equilibrium example.
Computer Vision is a game-changer in the real estate value chain and the inflection point is upon us. The question has gone from theoretical to practical as the pace of innovation accelerates during this period of intense change.
So what is Computer Vision, also known as CV? Powered by Machine Learning, Computer Vision is the art of teaching computers to see and interpret the same way humans do, whether it’s photos, books and documents, a dancer’s moves or the condition and desirability of a house’s interior. It has been made possible by advances in and reduced cost of computation, the ubiquity of devices and sensors, cloud-computing, and innovation in software algorithms. Computer Vision at scale is only really possible now.
Well and good, one might say, but what does this have to do with real estate? Let’s start with a subject at the heart of the industry: valuation. To understand how Computer Vision can radically change the valuation industry, a brief tour of existing automated valuation model strategies is valuable. Current AVM models fall on a spectrum between two extremes. On the one hand, we have pure hedonic models that take into account the ordinary property characteristics, such as the square footage and year built. Posed against that are the comparable property models that value a house based on the average of comparable house prices sold in the last couple of years in a marked radius from the subject property. Neither of these models captures the condition of either the sold properties or the property being valued.
This is where Computer Vision comes in. MLS listings typically have about 20 interior and exterior photos of each listed house. With Computer Vision, these photos can be quickly analyzed to detect and classify these photos (which is a kitchen, which is a bathroom, etc.) and subsequently, each room type is rated by a specialized image classifier and given a score or a signifier (modern, average, dated, etc.).
Computer Vision “object detectors” can then be used to detect individual features like double-wall ovens, dated electric stoves, waterfall islands and other “details,” which, to a buyer or a seller, are incredibly important and can affect valuation.
With this detailed level of condition data, an AVM can do fine-grained adjustments, as well as perform more accurate comparable property selection and valuation.
Computer Vision doesn’t only improve AVMs, it also enhances other components of the real estate value chain. A few examples:
- Portfolio analysis: Imagine being able to analyze a portfolio of thousands of properties based on inspection or MLS photos to get a condition score for every property. Now imagine scaling this up to the 100+ million residential plots in the U.S.
- Appraisals: Image analysis can be used to help either automate appraisals or to flag properties that appear distressed, high-end or well-maintained.
- Visual property search: Clustering properties by visual similarity would allow one to search properties by visual characteristics. For example, it could answer the question, “Show me other properties with kitchens similar to the one in the house I just saw?”
- Satellite or aerial imagery analysis: It can help find which properties have solar panels, pools, outbuildings, or other visible characteristics.
- Remodel analysis: Since Computer Vision can provide the condition of rooms such as kitchens and bathrooms, it is possible to do detailed ROI analysis on questions like, “Should I upgrade the cabinets or countertops before selling?” or “Do other homeowners in my area upgrade their bathrooms before selling?”
These are just five examples of Computer Vision at work, but the extensions and enhancements are endless.
Residential Real Estate is the world’s largest asset class. In the U.S. alone, it is estimated to be between $37 trillion and $40 trillion. More than that, it is a bellwether sector that affects the entire economy. Both booms and recessions occur as a result of changes in the real estate market. As such, the industry needs to pull out all stops to the application of the latest technologies to the marketplace.
Despite the enormous potential and fertile possibilities, Computer Vision isn’t perfect. Inconsistencies in labeling, imperfections in image detection, or even a lack of a critical mass of data for the “machine to learn” are issues still being worked on. The rate of innovation and improvement in Computer Vision is exponential so any Computer Vision solution should be scrutinized for extensibility.
Computer Vision is ushering in a new era in real estate. All industries change, some more rapidly than others. The real estate market is primed for its most exciting phase ever, and Computer Vision will play a huge role in this wonderful process.
For the full June HW Magazine issue, go here.
