The current hype around artificial intelligence (AI) has many in the real estate industry dreaming about harnessing it to assist with investment decisions. But there’s a giant stumbling block: too much bad data.
AI models require voluminous data to be accurate, and commercial real estate data is not standardized or consistently high-integrity. AI adoption in real estate investment is “in diapers,” says Matias Recchia, CEO of Keyway, an AI-powered investment management company focused on sale-leasebacks—part of the Camber Creek portfolio.
To apply AI to investment-related data, you have to “normalize” the data, says Kenter Wu, CTO of Dealpath, one of the pioneers in bringing AI to the bargaining table. He explains that by removing anything unstructured, redundant, or low-integrity and creating a logic-based way of storing it, “you’re looking at apples to apples.”
According to Rick Haughey, who consults with real estate companies on their digital transformations, the industry “doesn’t have a lot of consistent data, and [it is] not great at making data consistent and cleaning it up.”
The “fragmented data” dilemma
Data fragmentation poses a significant challenge to AI in real estate investment, with companies juggling multiple, non-communicative software platforms.
“We have a corporate member of our organization that collects data across their company on their properties on 40 different software platforms–40!–and none of them communicate with each other,” says Lisa Stanley, CEO at OSCRE International, a nonprofit organization focused on the development and implementation of real estate data standards. “I don’t think that’s the exception. I think that lack of communication is more the rule.”
Before they can effectively use AI, investors have to “pull the data from all these different systems and have confidence in the data quality,” she says. “Even with AI, garbage data in still yields garbage data out.”
However, not every company has access to every data source. Most real estate data is privately held, and that’s by design.
“A lot of [investment] managers work off of asymmetric information,” says Andrea Jang, founder of AJS Advisory, which advises real estate startups and investors on organizational strategy, growth, and operations. “That’s how they get faster access to a deal…a tip…[that] may not be available to everyone.”
In the tech world, there’s a unifying belief that “all data should be accessible,” says Jang. “It’s not like that in real estate.”
Unless federal, state or local jurisdictions otherwise mandate that specific data be released, investment professionals “are going to keep as much data as they can to themselves in the private equity markets,” he says, “because that’s what makes one investment manager better than the other at the end of the day.”
Harnessing AI takes long-term commitment
According to Jang, many tech companies and tech teams within large real estate firms are focusing a lot of their resources—human resources and capital resources—on structuring and standardizing data.
It’s not a simple process.
Faropoint, a tech-enabled real estate asset manager, began its data efforts roughly four years ago in preparation for training its AI models, according to CEO Adir Levitas. It involved hiring a research and development team of data engineers, which now totals 20 people in a company with 120 employees.
“Fifteen to 20 percent of our headcount will always be tech people,” says Levitas. Effective use of AI for commercial real estate investment “is a long-term commitment to modify your organizational chart. [It] starts with data infrastructure and architecture.”
One of Faropoint’s analytical AI models creates a linkage between any address and market rents on last-mile warehouses, which requires “hundreds of thousands of data points,” according to Levitas.
To build the model, Faropoint accumulated a “ton” of data and developed a proprietary system to ensure the data were of high integrity. Then, the team had to determine which parameters were most likely to affect rent, such as age of the property and proximity to public transportation. Based on the findings, Levitas says the list of parameters was narrowed down from hundreds to a couple of dozen.
Levitas stresses that the work doesn’t stop when the model is built and functioning. It needs to be consistently updated.
“Creating an accurate AI model is 30 percent of the mission,” he says. “Seventy percent of the mission is to integrate AI into the business processes that are going to be useful for people. It is a day-to-day process that starts with buy-in at the top [and] a long-term commitment to fine-tuning, [receiving] feedback, and [being] willing to [find] biases and noise in the system, which will be there because we’re human.”
The push for real estate data standards
Widespread data standardization could enable greater AI usage by more companies, says Stanley.
Standards like those in OSCRE’s Industry Data Model™, which is open access, can help those individual vendors speak a common language.
“Data standards are a base layer of common terms and definitions and facilitate data exchange across platforms,” Stanley says. “One layer up from that is performance standards, such as LEED certification. A layer up from that is data management, governance, and strategy.”
She says that greater standardization could get a push from new SEC reporting requirements around greenhouse gas emissions, as well as similar measures from state and municipal governments, that require companies to get their data in order.
People will always have final say
While some firms are making significant strides toward using AI to augment their real estate investment decisions, in no case has AI replaced the human touch.
Recchia views Keyway as the “Bloomberg Terminal for real estate,” with AI implemented to make buyers, sellers, and brokers “more informed with the same data on both sides of the deal” and “able to analyze faster, saving them time.”
AI models outperform humans in data analysis but miss nuanced factors, Jang explains.
“At the end of the day, there is a lot in real estate that is still very subjective,” she says. “A human can visit a site and very quickly determine whether it’s right for an acquisition or a development. Training the AI to do what the human eye does naturally is incredibly laborious and expensive.”
Real estate firms should conduct a cost-benefit analysis before investing in AI. But in an industry that has been historically slow to adopt technology, they shouldn’t sit on their hands either.
“A lot of folks should be using [AI] for asset management, but they are instead [focused on] their problems around interest rates and are not prioritizing tech,” says Recchia. “But the technology is out there, and it can help them save costs and make better decisions.”
Case studies: AI’s nascent rise in commercial real estate
Artificial intelligence has evolved in recent years from being solely analytical—so-called “traditional AI”—to the more well-known “generative AI” popularized by ChatGPT and other platforms. While investment in AI as a profit-making decision tool is still in its infancy, several forward-thinking firms are pioneering its use with both traditional and generative AI models. Here are some of them:
Dealpath: The company incorporated traditional AI into its dealmaking platform in 2020, long before the current AI boom, according to CTO Kenter Wu. Today, the platform uses Dealpath’s proprietary traditional models along with generative AI components.
The programs extract large volumes of data from PDFs, Offering Memorandums (OMs), Excel files, and even paper flyers, and do it far more rapidly and accurately than people can. The platform then sorts the data into tables, making it easy for investors to see relevant deals by using filters, saving them “tons of time and tons of effort,” Wu says.
Faropoint: According to Adir Levitas, CEO of Faropoint, the asset management firm has created several analytical AI models, the most advanced of which predicts the market rent for industrial real estate for any given address in the U.S. (Leasing data is an essential filter because last-mile industrial warehouse owners make money by re-leasing properties to the market.)
Faropoint is using the model today in an “advisory role,” Levitas says. With its ability to underwrite deals and provide ballpark rents at any location, he thinks of it as a member of an investment “committee” where it tees up various comps for human investment professionals to parse through, saving valuable time in their decision-making process.
“We eventually sent people to verify,” he says, “but we were able to put in an initial offer of hundreds of millions of dollars on a large portfolio just based on that initial [AI] underwriting.”
JLL: According to Richard Bloxam, JLL’s global CEO of capital markets, the firm’s investment advisors use a platform called Horizon AI. This platform is equipped with proprietary machine learning models and AI algorithms that combine JLL’s own data with third-party data to help identify, analyze, and source opportunities. Following an investment from JLL Spark Global Ventures, the corporate venture arm of JLL, the company wholly acquired AI-powered commercial real estate investment management platform Skyline AI in 2021. Horizon AI was developed by combining Skyline AI with JLL’s existing quants “to deliver faster insights for real estate experts to act on,” Bloxam says.
“It is a powerful application that, in the hands of our experts, allows them to better advise clients on investment opportunities across acquisition, disposition, and lending opportunities,” he adds. “In fact, in 2023, one-in-five of our capital markets pipeline opportunities globally was enabled by this platform.”
Keyway: The investment management firm Keyway uses traditional AI to source acquisitions, according to CEO Matias Recchia. Its platform targets the middle market, where good data is harder to find, and ingests deal data and analyzes acquisition opportunities. Recchia says Keyway is licensing the platform to third parties, including a pilot with multifamily asset managers.
“We are testing features with generative AI and our own language learning models,” he says.
“But right now [the platform] is analytical AI with some predictive [capabilities].”
Interested in learning more about what experts have to say on AI and CRE? Become a member for exclusive access to the insights discussed in our upcoming member-only webinar on May 29th, 2024. Stay up to date on other trends shaping the industry through our diverse array of topical reports, global and local networking events, monthly webinars, learning courses, and more! Register now!
