The Genesis of Generative AI for Everything Everywhere All at Once in CRE

reprints


Earlier this month, global commercial real estate services giant JLL began rolling out JLL GPT, which it claims is the first bespoke generative artificial intelligence (AI) purpose-built for commercial real estate.

Developed by JLLT, the company’s technology division, JLL (JLL) GPT draws on in-house and external CRE data. It is intended to be used not only by brokers, but also by the firm’s more than 103,000-strong global workforce.

SEE ALSO: HqO’s Chase Garbarino On the Intersection of Proptech and Human Resources

Generically, GPT is the common term for generative pre-trained transformers, which are a family of neural network models that uses the transformer architecture, making it a key advancement in generative AI applications such as ChatGPT.

“Generative AI is a pretty new technology for commercial use,” said Yao Morin, who as chief technology officer at JLL heads the JLLT team. “Since generative AI has been broadly adopted, we have been working on exploring how it can be used in JLL. And JLL GPT is just the start of how we use generative AI. But we actually have quite a few different POCs [proof of concepts] that we are looking into.”

JLL has already deployed AI technology to improve building efficiencies, generate 3D leasing visualizations, calculate sustainability risks and power investment leads, according to the company, which noted that 20 percent of all JLL Capital Markets opportunities globally were enabled by its AI-powered platform in the first quarter of 2023. The company plans to offer made-to-order solutions to clients later this year.

“Connecting buyers, sellers and lenders at the right time, with the right data in hand — within seconds — is going to determine success in this new generative AI era,” Richard Bloxam, JLL’s CEO of capital markets, said in a statement. “We are already using AI to optimize investment opportunities, creating a competitive edge for our clients. Leveraging comprehensive real estate data with JLL GPT’s insights trained by our capital markets advisers will significantly enhance decision-making.”  

The essence of JLL GPT involves infusing the company’s in-house data and general CRE knowledge into a generative AI large language model, said Morin.

“It has a chat-like interface which is enabling the entire JLL workforce to use that interface to actually ask questions and to do different kinds of work that they need for their day-to-day operations,” she said. Since its launch, the GPT has 12,000 unique JLL in-house users, Morin added.

JLL is happy and somewhat surprised by its employees’ early adoption of GPT, but like all such technology the platform is just beginning to iterate, Morin said.

“We take a pretty typical software development approach where we want to release a minimum viable product so that we can start getting user feedback,” said Morin. “That’s how you want to develop a good product. It’s not like you close your door and hide in a corner to develop the product. You want to get user feedback as soon as possible.”

JLL employees are providing “tons of feedback” via daily emails on the GPT, she said. “It’s going to be a product that will continue to evolve. 

“We see some people who ask JLL GPT questions more than 100 or 200 times a day,” said Morin. “And then there are people who ask us, ‘Can you show me examples of how to use it?’ So we are gathering use cases to show them. There’s different acceptance levels just like any other technology, but we are very excited about the feedback that we have been getting. The engagement is amazing. It’s actually very energizing for my team and myself to see that kind of engagement.”

So who answers these employee comments, JLL GPT or a technologist?

“We actually have been manually responding, but the funny thing is we have received some feedback that is generated by JLL GPT,” she said. “The employee uses it, has some feedback, and then asks the GPT, ‘Can you provide feedback for me based on these three points?’ Then JLL GPT will generate a really nicely written email for them and they send it to us. It’s really interesting.”

One of the most common and painfully mundane issues the GPT addresses is emails between JLL professionals and clients, said Morin.

“Usually it takes a lot of time to write those emails,” Morin said. “Brokers describe to us needing to rely on prospecting emails to a particular company, which usually will take him an hour or so to write. Using JLL GPT he was able to do it in 15 minutes.”

Initially asking the GPT to produce a prospecting email yields a generic product. But by further prompting the AI with questions pertaining to JLL’s capabilities, a client’s recently announced sustainability initiative and other specific inquiries, the final product can be refined and personalized, she said.

“That kind of situation is what thousands of brokers or other JLL employees do every day now,” said Morin.

JLL GPT is also being used to read, summarize and even create tables on demand from the vast array of CRE documents that flow through the company, many of which contain obscure or incomplete language even for a human reader.

“Here’s a funny example of how you don’t need to manually extract the information that you need,” said Morin. “It’s hard to find out the pet policy in a lease. Sometimes you want to know, ‘Hey, can I bring a dog to the office?’ You know how difficult it is to actually find out? Usually they don’t say anything specifically about a dog. It will be hidden in a document. Now you can just ask JLL GPT, ‘Hey, can I bring a dog to the office? This is the office address.’ Then it can tell you whether you can bring one.

“You don’t even need to translate that to, ‘What is the pet policy of this office?’ You can just say, ‘Can I bring a dog,’ and it can understand what you’re asking and pull out the pet policy. So that is a game changer.”

As nimble as its technology may be, though, JLL is concerned about transparency in its use. The company has AI responsible use guidelines, such as noting that a product has been created with the use of AI.

“You cannot use anything generated from an AI tool directly,” said Morin. “It needs to be a starting point, not the final product. Second, if it’s written by JLL GPT, they need to credit it just like you would credit anything if you quote something.”

Despite JLL and Morin’s pride about JLL GPT’s deployment, she declined to disclose any information about its development, including the time it took to create, how many people are on her team, or who worked on the technology.

“We do need to keep some secrets as this is a very competitive field,” Morin said.

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