When Gabe Marans negotiates a lease for one of his tenant clients at Savills, he expects pushback on construction allowances or maybe rental escalations. What he doesn't expect is a 3,000-square-foot tenant in a three-million-square-foot building demanding lobby signage rights because an AI chatbot convinced them it was standard. Yet that's exactly what happened; not once but multiple times in recent months, throwing deals into chaos at the worst possible moment.
The vice chairman at Savills described the phenomenon during a recent appearance on The Not Podcast, hosted by CompStak Co-Founder and CEO Michael Mandel. After working through term sheets, engaging outside counsel and preparing final lease drafts for client approval, Marans has watched clients return with AI-generated comments that fundamentally misunderstand market norms. The result forces brokers into an uncomfortable position of de-educating clients who arrive at the negotiating table armed with confidence but lack context.
The Misinformation Problem
The issue surfaces after lease negotiations have progressed significantly. Savills shares comments with the landlord, outside counsel contributes their review and the revised draft goes to the client for final approval before submission. That's when clients consult AI tools and return with demands that reveal a gap between algorithmic output and market reality.
"We're put in this position where we have to de-educate our clients in a way that we never had to before," Marans said during the podcast. These AI-driven requests have moved beyond silly examples to major issues that put teetering deals at risk during the eleventh hour.
The lobby signage example illustrates how AI tools can mislead clients about what constitutes reasonable lease terms. A tenant occupying roughly one-tenth of one percent of a building's square footage has virtually no leverage to demand lobby signage, yet AI told them otherwise.
Turning AI Into An Asset
While Marans sees clients misusing AI in lease negotiations, his team at Savills has developed more strategic applications. He's trained an assistant tool, Claude, using seven to eight years of his LinkedIn writing to understand his voice and content, creating what he calls an online editor. Before posting anything publicly, he runs it through Claude to check whether he's covered the topic before, confirm he isn't contradicting previous posts, and refine the length or language.
The firm has also connected Claude to email and calendar systems to analyze response patterns. Marans uses the tool to track how long he takes to respond to emails, whether he favors certain contacts over others and how his behavior differs between outside clients, outside brokers, and internal colleagues.
"We're using it to do some pretty detailed analytics," he explained, noting the goal is identifying gaps in their business plan and areas for improvement.
CRM Integration And Lease Review
Savills has incorporated AI into its Salesforce CRM to flag relationships that risk slipping through the cracks. The firm isn't using automation to send mass emails, which Marans prohibits on his team. Instead, the technology proactively surfaces data that helps brokers determine who deserves outreach without removing the human judgment required before hitting send.
"It's really important for us not to let things slip through the cracks, and at the same time not to make it automated," he said.
For lease review, Marans uses Claude to confirm information and spotlight provisions that deserve closer attention, though he emphasized the technology can't replace manual review. A broker who relies on AI for lease analysis and misses a hallucinated provision risks losing a client for life. The tool functions as a supplement that accelerates document review rather than a replacement for expertise.
Three members of Marans' team are currently experimenting with building agents on Claude to improve workflows and maximize efficiency. He acknowledged the learning curve, noting "we're going to fall before we can walk," but sees the experimentation as central to how the team will evolve its practice. The entrepreneurial culture encourages junior brokers to test applications that might eventually become standard across the firm.
The Broker's Dilemma
Marans finds himself caught between two realities. On one hand, he wants clients who show up informed and engaged.
"We want our clients to be empowered, we want them to be knowledgeable, educated," he said, explaining that repeat clients who understand the basics make his job easier. But the information they're pulling from AI rarely includes the market context that separates a reasonable ask from a ridiculous one.
The problem has gotten worse as more people flood the market with outreach. When Marans started brokering deals, cold calls converted at three to five percent. Now that the rate sits closer to one percent, which means brokers need to work harder just to keep their pipelines full. At the same time, clients armed with AI-generated lease reviews are slowing down transactions by raising issues that wouldn't have come up a few years ago.
Instead of moving deals forward, brokers spend time walking clients back from positions that made sense to an algorithm but do not pencil in practice.
What Marans has figured out is that AI works when it supports what brokers already know, not when it tries to replace that knowledge. His team uses it to track email patterns and surface overlooked relationships in their CRM. Those applications make them faster and sharper. But when it comes to actually negotiating a lease, the technology still can't match the judgment that comes from years of watching deals succeed and fail.
© Arc, All Rights Reserved. Request academic re-use from www.copyright.com. All other uses, submit a request to TMSalesOperations@arc-network.com. For more information visit Asset & Logo Licensing.