AI for Real Estate Documents: Are Lawyers’ Days Numbered?

Unsurprisingly, the tedious business of real estate legal document reading, vetting and abstracting is becoming easier, faster and more accurate through greater use of artificial intelligence.

However, the road to reducing lawyers’ roles in untangling the Gordian knot of real estate documents remains bumpy, even as landlords, developers and brokers steadily, if somewhat slowly, embrace AI to speed workflow and to lower costs.

Meanwhile, generative AI continues to accelerate seemingly at the speed of light.

One proptech startup that is using AI for contract reading in the construction industry is Superlegal, a Tel Aviv-originated company that began three years ago, said Noory Bechor, Superlegal’s CEO and co-founder.

“We work with construction companies and subcontractors helping them review and negotiate contracts more efficiently using AI,” Bechor said. “There are other technologies that help you, for example, scan through documents and identify things that are important for managing a project, like delivery schedules or specs, but our focus is basically creating an alternative to hiring a lawyer or choosing an outside counsel. That’s our unique value proposition.”

Superlegal’s United States headquarters is in Lehi, Utah, and it operates throughout the country, working mostly with small and medium businesses.

“Utah was the first state that came up with an initiative to create more competition in the legal market and to offer new alternative solutions to hiring law firms,” said Bechor. “The goal of the regulator there was basically to create more competition in the legal market. They understood that the legal market is not competitive enough because law firms are established in a certain way, and there’s basically a problem of access to justice. So, sometimes small and medium businesses, or even consumers, can’t afford lawyers, or decide not to use lawyers because they’re too expensive.”

The goal in Utah, then, “was to use technology to level the playing field and give these guys access to new tools that are cost-effective and fast, but can help protect them — just like the large guys have protection — by having lawyers on their team or using an outside counsel,” Bechor said. “We became the first AI company in the world to get a special license from the Utah Supreme Court to actually give legal advice and practice law as an AI company, as a tech company, not as a law firm. So that’s why we are headquartered in Utah.”

Superlegal has a team of lawyers in Utah that uses the company’s AI to supervise and verify the technology’s results. “It’s one of the unique things about our solution, in that it gives you the benefit of AI, but with the supervision of an actual human attorney.”

Along with saving time, AI can replace the outside counsel used to review a construction contract, which normally costs several thousand dollars per contract. “With Superlegal, we’re talking about the low hundreds of dollars, like $300 per contract,” said Bechor.

Another proptech startup tackling legal work for real estate is Bryckel, a company based in tech-heavy Mountain View, Calif.

“We’ve built AI for real estate document intelligence with legal grade precision,” said Sienam Ahuja, CEO and co-founder of Bryckel. 

She said Bryckel’s model can do three key things with real estate documents.

“It scales due diligence through a proprietary product called Clause Protect, which analyzes any lease across any asset type to give a comprehensive risk report,” Ahuja said. “The second thing we scale is deal transaction velocity. So, for instance, a letter of intent (LOI) to lease comparisons can be very intensive, because you want the business terms to be reflected in the lease draft every single time that it is red-lined and shared across parties. We are able to demonstrate that everything in the lease has reflected what is in the LOI.”

Finally, Bryckel can scale operational efficiency across both leases and the entire knowledge base, said Ahuja. “Property managers shouldn’t be coming and asking: ‘Hey, the tenant wants to sublease, but I don’t understand the sublease clause,’” she said. “It’s in conversational English for them to understand and be able to handle rather than burden the in-house teams.” 

Using proprietary artificial intelligence on top of foundational AI platforms like Anthropic, Mistral and OpenAI, BryckeI is designed primarily for asset managers, brokers and property managers in mixed-use assets. Legal teams can also use Bryckel, but the idea is to reduce their workloads, said Ahuja.

Bryckel spent a year validating its entire AI pipeline to make sure that the information was accurate, and is now at 99 percent or more accuracy, said Ahuja. “Precision is one of our cornerstones, which is why you would use us rather than just throwing your document into ChatGPT.”

Founded in 2018, San Francisco-based Prophia is a more mature proptech startup that originally focused on technology for lease abstraction, said Cameron Steele, its co-founder and CEO. He said the growth in innovation in the seven years since starting the firm “has just been staggering.” 

“We have three full-time data scientists, so about 10 percent of our employees are working on this problem,” Steele said. “They even have trouble keeping up with some of the underlying innovation that we’re leveraging. It’s moving fast and furious. I would say that in the last nine months — largely driven by the large language models (LLMs) — our ability to read, transcribe, improve, and then develop on top of some of these core technologies has really been enhanced.”

That’s opened the door to measurement statistics. “We measure the enterprise-grade quality of output of what we do, the human review, and then talk about risk versus who takes on risk of output,” Steele said, “because that’s a really important area that our customers care about.”

While prophia still focuses on lease documents, its longer-term goal is to expand AI review to enterprise-grade accuracy levels for other document types, Steele said. That includes loan documents, service agreements, joint venture agreements and anything else pertinent to transactions and managing commercial assets.

Prophia’s AI is layered with human review, which has increased its annotation accuracy from the 80th and 90th percentiles to 99 percent, he said.

“We used to do training models on top of what I would call small language models, or natural language indexes, and now it’s all generative,” said Steele. “We’re just literally doing prompt engineering and workflowing, and then tweaking based on the results we get from some of the engineering work we do on top of these models. I think over time there will be the ChatGPT kind of low-end, pretty good version that a lot of people could use and benefit from, but then there’s enterprise level with high expectations on accuracy and output.”

AI changed radically about two years ago, said Vijay Anand, vice president of AI at MRI Software. He spent 20 years at Ernst & Young, now EY, leading AI product development on the firm’s global innovation team before ChatGPT arrived on the scene.

“Document extraction is a specific technology,” said Anand. “At MRI Leverton — which we’ve rebranded as contract intelligence, or MCI — MRI contract intelligence has been around as one of the pioneers on extracting data from contracts, amendments and any kind of real estate-related documents, running the gamut from a maintenance agreement to any kind of addendum.”

The AI world in general, including real estate documents, changed with the release of LLMs that require huge amounts of computing power, Anand said. 

“These super large models are built on billions of data points that OpenAI pioneered and released,” he said. “That was the pivotal point in all our AI work today, where, before, the deep learning models were very specific to a particular domain and a subject. So that big change on these large language models was the pivoting point for not just MRI, but all of the vendors using AI to extract legal documents.”

The result is that MRI believes it has a technical advantage over proptech startups in this area.

“Legal contracts are such a broad domain, and real estate and proptech is a very narrow domain,” said Anand. “Expertise of your legal administrators and lease specialists becomes important in fine-tuning these models or even doing the right prompting. That’s where I think we have an advantage, having done more than two million leases. So, experience and human expertise is still important, even in this age of AI and LLMs.”

Investors are another group keeping a close eye on legal document AI for real estate, said Jake Fingert, managing partner at proptech venture capital firm Camber Creek.

“I don’t expect, in the near term at least, for AI to get rid of legal fees altogether,” said Fingert. “I think there is still a lot of really valuable work for lawyers to do, particularly lawyers who understand kind of the art of the deal, to help advise real estate owners and operators. I think there’s a lot of value there, but I think a lot of the lower-value work can go by the wayside, and real estate firms can save a lot of money with some of these AI tools.”

In the universe of emerging AI companies, startups include Bowtie and Cadastral, which work on technology strictly but broadly for the real estate industry, as well as legal tech firm Harvey, which has a real estate vertical, said Fingert. 

“Those firms focused exclusively on real estate are doing a range of things — from analytics and business intelligence to extraction and anomaly detection,” said Fingert, referring to Bowtie and Cadastral. “They’re even helping to draft documents, and, in some cases, will help with the negotiation.”

Investors have options as to where to place their bets, but market choices are definitely a gamble.

“We are certainly interested in those types of solutions,” Fingert said of generalist and real estate-specific AI companies. “One of the questions that we’ve been thinking about a lot internally is whether there is a big enough opportunity for real estate-focused legal tech companies to really become big billion-dollar companies. Or, is the more interesting opportunity to invest in the platform companies and then to embed that as a module, or even work with one of these generalists? Invest in one of the generalist companies where they’re working across different verticals, and they have a focus on real estate, but given the scale and the breadth of their offering, they’re able to operate at a different level than some of the real estate-focused companies.”

Andrew Zang is a real estate attorney, as well as the CEO and managing partner at Bespoke Air AI, a Manhattan-based startup that he formed out of his experience and frustration with the real estate paper chase.

Having seen the early days of using ChatGPT while working at Savills, Zang said he became aware of the lack of confidentiality and privacy safeguards with the technology.

“So we asked, ‘Who is looking over this foundational model that’s been built by ChatGPT?’ And the answer was, just as it is in many companies right now, they’re really interns, because there are no professional teams with subject matter expertise that are being paid to review the output that these systems are producing.”

Bespoke Air is designed to provide proper AI oversight for real estate clients by avoiding foundation models like ChatGPT and curating AI solutions for the individual client company, said Zang. For instance, landlords can AI-automate their building stack, market reports and lease data through Bespoke Air’s retrieval system.

“What we do is artificial intelligence review,” he said. “We’re the managers of the risk and the responsible AI component of everything. We ensure accuracy by doing three months to six months of edge testing on all of their leases and materials to make sure the AI doesn’t hallucinate. We make sure that if you ask it what you had for dinner, it doesn’t answer the question.”

Having started Bespoke Air in January, Zang is still dealing with sluggish industry adoption.

“Do the landlords get it?” Zang said, repeating a question. “The sons do.”

Philip Russo can be reached at prusso@commercialobserver.com.