For the last several years, the residential real estate industry has been abuzz with the phrase artificial intelligence (AI). Seemingly every technology company has marketed it as a key part of its offering.
While the concept is widespread in the industry and components of AI exist in many products, AI’s actual application and status in the industry may be different than many think. As such, this article explores the current state of AI in real estate.
The idea of AI applications range from simple programmed response to thinking machines. Programmed response is just a form of programmed automation and actually is not AI. Applications include if-then-then-that type automations. Thinking machines, on the other hand, are the ultimate in AI, as Sean Gourley, CEO of AI firm Primer AI, spoke about at this year’s T3 Summit.
The type of AI Gourley outlined dealt with deep learning, AI systems that did not need training but could develop insights themselves by analyzing raw data. This is not the type of AI in real estate today; real estate AI currently is directed with an extremely organized set of data.
AI in real estate
The utility of all AI in real estate centers on training with structured data. The system learns to detect patterns, connections and relationships from that training data in a process known as “machine learning.” It applies this learning to similar data it has not seen before to generate responses that lead to insights or automated responses.
A group of potential outcomes is assembled from the training material, and confidence ratings are assigned depending on how closely the new situation matches past situations.
|Real Estate and AI |
AI: Automated pricing algorithms used by companies like Zillow Group, Opendoor, Redfin and many others. Systems that detect features based on pictures of the property. Suggesting properties based upon customer search histories and profiles. Predicting closings based on recent trajectories of similar properties. Systems that suggest marketing plans based on training of similar properties, markets.
Not AI: Workflow automation between different systems such as those that “read” documents and automatically input text into digital documents downstream. Most chatbots, who typically pick up on certain keywords or respond via preprogrammed conversation trees.
But because training data plays such a large role in the efficacy of AI applications, real estate technology has so far been relatively limited. For example, a system that answers the question, “How long will this property take to sell?” or “What will this property sell for?” requires a history of listing activity for similar properties to generate an answer.
And, because the outcome is only as good as the universe of data that goes in, the system needs access to local and national economic data, that of the local market over time, changes to area zoning and more to generate an accurate answer. Assembling custom data sets (listings or consumers) with other difficult-to-obtain data points is what makes AI so difficult. IBuyers are using machine-learning AI to supplement their modeling for offers, pricing and sales.
The large number of historical events needed to support machine learning is a drawback to adoption. Standards for transaction events are being discussed and prototyped in the RESO standards group, but no publication has been made yet.
|The cadence of AI development in real estate: Current: Image recognition and market pricing. Nearer term (2-5 years): Marketing plans and prospecting. Longer term (5-10 years): Chatbots that intelligently and appropriately respond to consumers.|
Because the magic of AI is the training base, images offer a great option for AI applications, as they contain much information, much of it standardized. For example, a system can be trained to recognize an ottoman in a photo pretty easily – there are tons of photos of ottomans of all types and sizes the system can be trained on.
Machine learning AI is being used with property images by MLSs in many markets with the help of companies such as RESTB.ai, REAI and Vize.ai. The following examples of image processing include:
- Detection of image violations
- Creating image descriptions for ADA compliance
- Removing watermarks from images
AI could expand to property listings if the burden of creating training data is reduced. Standards are one way to reduce training efforts because multiple sources of data could then be used to create a larger pool of information for AI systems to leverage.
AI has the opportunity to greatly advance the real estate industry, but it is still in its early stages. Companies in the industry cannot yet realize many of the promises of AI as data standards and integrations have yet to progress to a point sufficient to power the training necessary. However, progress is being made. Stay tuned, as we at T3 Sixty are tracking AI in real estate’s development closely.