Commercial Real Estate Owners Are Just Starting to Decode AI

Commercial real estate owners can’t be blamed for being leery of using artificial intelligence in their organizations, as accuracy and trust issues still plague many attempts to grab hold of a technology whose evolution appears to be rocketing forward daily.

Many proptech entrepreneurs and investors who themselves admit that keeping abreast of AI’s iterations in real estate is a daunting task have strong opinions about how CRE owners need to address the issue and how they are understandably lagging in doing so.

No matter what angle the proptech experts come from, they all agree owners are struggling to understand how, and with what, AI providers can meet their greatest needs: accuracy and trust.

“The industry has heard the word AI for the better part of the last three years,” said Prasan Kale, co-founder and CEO of Outcome, which uses AI to automate CRE workflows and data. “It’s everywhere and it’s a big word. It’s not one thing. It’s many, many different things. It’s a methodology more than a product. But what’s happened in the hype cycle of AI is that owners felt, ‘Oh! There’s this magic wand called AI that can come in and solve the problem for me.’ That can’t be further from the truth.”

Some proptech companies tried “putting a wrapper on ChatGPT” and selling it to owners, but such pilots failed because they were based on generalist AI rather than CRE-oriented AI, Kale said. That was more a point solution rather than end-to-end workflow-based, he added. Kale, along with his Outcome co-founder and Chief Technology Officer Sid Jain, also founded Rise Buildings (later acquired by VTS). The duo brought the Chicago-based Outcome out of stealth mode last week.

A couple of themes emerge when Aaron Ru, principal at RET Ventures, a venture capital fund backed by more than 50 institutional real estate owners, discusses AI with his investors. One theme surrounds who controls data. 

“How do you set an institutional data strategy to own more of your data?” Ru said. “The downstream effects of owning your data is that it allows you to train your internal AI systems and leverage data points you’re collecting across multiple systems such as property management system, internet listing service, email, chat, etc.”

In addition, centralization is key to getting benefits from AI, said Ru. “Consolidating leasing, renewals, maintenance and back-office functions really helps with AI implementation.”

Ru agreed that accuracy and trust remain major challenges for owners.

“Many of our portfolio companies that have successfully deployed AI in a proptech environment have done so with co-pilot-type deployments, where they work in conjunction with humans to help mitigate this issue,” he said. “As an example, our portfolio company Lula has deployed a co-pilot alongside their employees to help streamline responses to customer inquiries during the maintenance process. This has substantially reduced employee time spent per interaction. With the training data, they are now able to commercialize this co-pilot and offer it to external customers.”

While AI has to be embedded in owners’ workflows to be most effective, it also has to fit in with any security and compliance standards within those workflows, said L.D. Salmanson, co-founder and CEO at Manhattan-based Cherre, one of proptech’s fastest-growing data companies.

“The first thing is, does this do something valuable for the most part?” Salmanson said of clients’ concerns. “The answer is yes in theory, but not actually. It’s cool only if this is in a big process, embedded in my workflow, and passes security and compliance. As a stand-alone, I don’t need it. But if it can actually automate the monthly close, that’s value. I want that, if it is SOC 1 accounting compliant, and SOC 2 security compliant.” (SOC 1 and SOC 2 refer to compliance and auditing standards.) 

Another issue is that CRE owners whose companies are able to discern good proptech AI providers from bad are relatively rare.

“I wouldn’t even start talking to the company until I know that they’re at the level of enterprise maturity that can work with me,” said Salmanson. “I’m not even piloting, because even if I pass that gate, I’m never going to be able to deploy this in my organization. So it’s not even worth the conversation. I trust business users to know if a business process is of value or not. If I asked you, ‘How valuable is it for you to to abstract leases?’ you’d say it’s not changing your life. If I ask you, ‘How valuable is it to have your monthly close automated?’ you would say, ‘Well, that costs a lot of money.’ Or ‘How valuable is it to you to do your portfolio or accounting?’ You spend millions and millions, so you know intuitively what processes are worthwhile automating or not.”

The next question the provider has to answer for the owner is whether a process can be credibly automated.

“When an organization is asked how much judgment is being applied today by people in that process, if it’s very little judgment, it’s a very good candidate to be automated,” added Salmanson. “If that process has one spike in judgment, that’s a good candidate for a human in the loop process where it’s automated. If there are multiple such spikes, it’s probably not the best candidate.”

Vertically oriented AI proptech service company executives emphasized the importance of such tailored AI solutions over generalist platforms.

“It reminds me of Y Combinator, which sort of popularized the saying, ‘Don’t be a technology searching for a problem,’” said Andrew Thompson, CTO of Orbital, a London-based company that focuses on structuring and analyzing complex real estate documents using advanced optical character recognition technology. “That is a really bad idea to create a business, because you end up creating some amazing technology that is not very valuable.

“What are your business challenges? What are the competitive pressures at play? What are your clients asking you about? And sort of matching that to AI. I just think the game is slightly different over the last year in that the historic software was a lot more point-based solutions, especially when it comes to machine learning.”

Digitalizing real estate legal documents accurately with AI requires highly specialized technical and legal knowledge, said Thompson from Orbital’s recently opened Manhattan office. The company has grown from the United Kingdom and European Union markets to the United States.

“We’re talking about documents that are sometimes tens or hundreds of years old,” he said. “They’re very poorly photocopied, and so our customers will have these documents that are sometimes unreadable with the naked eye. Before you even get to the magic of AI, you have to get the data into a format. And, in real estate, you have the complexity of these documents that you can often have a lease with 10 amendments amending the lease over multiple years. So there’s a relationship between all those documents, and just throwing a bunch of PDFs into a generic AI tool doesn’t cut the mustard. It kind of falls over right before what the AI can do. So we spent a lot of time solving that problem.”

All of these challenges require productivity improvements and training, said Thompson, whose Orbital platform is used by various real estate professionals, including lawyers and title insurance companies.

Other proptech entrepreneurs such as Asaf Raz, vice president of marketing at Tel Aviv-based Agora, a real estate investment management platform that uses existing AI engines to streamline processes like membership agreements and tax document analysis, also emphasize the need for data quality over model quality.

“I would say the first issue for most of our customers, and also for us internally as a tech company, is that you can have the smartest model in the world, but have poor data,” said Raz. “That’s a big issue that we see on it all the time, and then the data becomes useless.”

Another prominent issue among AI adopters is wanting to immediately spread its use across the company, Raz said.

“If you have a workflow to focus on, make it clear before and after its use,” he said. “A lot of people don’t do that, and then they get tied up in everything that happens with AI, and it just doesn’t give them what they want.”

Grant Drzyzga, founder and CEO at Detroit-based Revela, a property management and accounting platform, said owners have to understand the need for diverse AI models to address various operational and analytical tasks. This is particularly true in putting the data in proper context and normalizing it for accurate financial insights.

“I think that everybody’s far behind the curve, quite frankly, us included,” Drzyzga said of owners’ skill with AI. “I think it’s changing and evolving. You have to bring in a very specific business problem to AI to solve it, because the things that these models can do, or the things that you can actually train your own homegrown models to do, are limitless.

“So, where we found a lot of focus and have made a lot of progress that’s been visible to early customers on our product is: What is the financial output you’re managing toward? And what are the building blocks that you can look at daily, weekly, monthly that gets you there? And, then: How can you have AI enforce standard operating procedures or look for anomalies before they become larger issues?”

Many CRE operators that are operating in older legacy systems, or that are dealing with a bunch of fragmented pieces of software, will struggle in this regard, Drzyzga said.

“The key to any AI implementation in a business is that the standard database is normalized and already relational, and is ready to have something analyze it ruthlessly, efficiently and quickly,” said Drzyzga. “Getting to that point is the hardest part, which is a little bit strange to think about. It’s almost like you have to dig the foundation of the building, and that takes a really long time before you start building upwards. But, as soon as you have that foundation dug, you can build sky high with a lot of these tools out of the box. You just have to have the right data infrastructure to do it, or it will be siloed, and it’ll give you the wrong results if you’re not feeding it all the context that it needs to be successful.”

Drzyzga also noted that while operators are curious about AI, they are skeptical due to high costs and the need for proof points, leading to larger operators investing in data and AI teams, despite facing challenges in building their own software.

Summing up CRE owners’ challenges in adopting AI, David Weissman, managing partner at East Brunswick, N.J.-based industrial developer Greek Real Estate Partners, emailed that for users to determine whether providers have real or exaggerated (read: useless) AI, “You must ask for measurable outcomes — reduced downtime, faster response times, or cost savings — and proof of integration with systems like Yardi or MRI Software. Avoid platforms that rely on buzzwords instead of performance metrics.

“In the short term, what’s needed is predictive maintenance and smarter workflows, while in the long term, portfolio-wide intelligence connecting energy, labor and logistics for true operational foresight.”

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