How JLL built an AI platform to help employees and clients close the deal

Navigating the post-pandemic commercial real estate market was quite a challenge for companies like JLL, as rising interest rates and work-from-home policies quickly changed the value of its holdings around the country. It required real-time access to public and private data, and early investments in generative AI helped internal and external customers find their way.

About one-quarter of JLL's 110,000 employees are using its internal AI platforms on a daily basis, said Yao Morin, chief technology officer for the centuries-old commercial real estate services company. Those users are researching deals, modeling cash flows, and developing applications for JLL's technology practice, and almost everyone in the company has used the platform at least once, she said.

"JLL recognized that this is our bread and butter in terms of, how do we differentiate ourselves from our competitors?" Morin said in a recent interview with Runtime. "If we can organize our data in a really meaningful and insightful way, that's a piece of our foundation."

The company was relatively quick to jump on the generative AI bandwagon, launching JLL GPT in August 2023 when many companies were still experimenting with the emerging technology. Built on Microsoft Azure, JLL GPT is a custom large-language model trained on decades of data on commercial real-estate transactions, and it was designed to help JLL's employees move faster when looking for deal opportunities and bringing them over the finish line, Morin said.

But as we've seen several times over the last few years, there's a common thread separating companies that have struggled to deploy AI from companies that have succeeded: early investments in modern data-management tools and strategies. Morin came to JLL in April 2020 — an anxious time in commercial real-estate circles — to build its data-warehousing strategy, putting Databricks' tools at the center of JLL's technology stack.

"JLL is a data business," Morin said. "Our clients are looking at us to provide insights to them, to provide things that only JLL does."

Commercial real-estate transactions are usually a matter of public record, but that data doesn't come out until weeks or months after the transaction is closed. But JLL's clients need access to market data such as the number of bids, the size of those bids, and how the market is responding when evaluating their own deals, she said.

The company needed a tool that could package that data for external clients while also elevating the internal capabilities of JLL GPT for deal-makers and software developers, which led to the launch of JLL Falcon in 2024, months after Morin was promoted to chief technology officer. Falcon allows users to access several different LLMs and provides more than 60 different AI-enabled features depending on the line of business tapping into the platform, Morin said.

"A lot of what we do is really looking at JLL being such a big corporation [and figuring out] how do we actually enable scaling, how do we enable speed, but yet making sure that there are guardrails that are built into our platform?" Morin said. 

"Every single one of my developers uses a coding agent."

Like a lot of companies over the last six months, JLL has deployed modern coding agents across its 800-person software-development organization. "Every single one of my developers uses a coding agent," Morin said.

The company tracks that usage on a daily and weekly basis, and measures not just the amount of code generated by those agents but the percentage that actually makes it into JLL's production code base. It is also allowing a "pilot team" to use agents to write every line of code deployed in those experimental projects; "the only thing that they can do is to teach the agent how to code in [that] environment," Morin said.

Coding agents are also forcing JLL's tech operation to rethink what it means to be a software developer in a world where increasing amounts of code are generated by the machine. Part of that process involves a lot of training and skill development, Morin said, and the shift is as much a mindset as anything else.

"I want our software engineers to become architects," Morin said. That involves teaching agents how to write code tailored to JLL's needs, but it also requires them to become "strategic planners" with a mandate to focus on the non-coding but absolutely essential parts of the software-development life cycle, such as data access, she said.