AI in CRE: From Pilots to Production, the Real Test Is Repeatability

In just three years, the number of companies piloting AI in commercial real estate has skyrocketed from 5% to 92%, according to JLL. But behind the headline growth, most organizations are still in the early stages of experimentation, with only a small fraction achieving their program goals.

This rapid adoption is colliding with a legacy of fragmented systems and inconsistent data. As firms race to integrate AI, the real challenge is not just getting the right answer, but getting the same answer every time—a standard that matters when investment committee decisions are on the line.

This article explores why repeatability and defensibility are now the critical tests for AI-driven underwriting, and how mid-market teams may be better positioned to adapt than their enterprise peers.

JLL reports that the proportion of companies running corporate real estate AI pilots has exploded from 5% to 92% in just three years. This surge signals a strategic pivot, with 88% of investors, owners, and landlords piloting AI and most pursuing an average of five use cases simultaneously. Yet, the same JLL survey found that only 5% of organizations report having achieved most program goals. "While 92% are piloting AI, only 5% report having achieved most program goals. Though implementation is widespread, most initiatives remain experimental with limited scaling," JLL noted in October 2025.

The experimental phase is not just about testing new tools. It's also about untangling years of legacy technology. JLL found that 81% of companies report at least three existing systems that aren't generating the expected results, and 88% are allocating budget to upgrade legacy technologies. Over 60% must address fundamental technology issues, such as duplicated functionality or dormant systems, before fully leveraging AI capabilities. This double burden—catching up on the basics while trying to innovate—means that even as AI pilots multiply, most organizations are still learning what works before scaling to full implementation.

For mid-market teams, especially in regions like Ontario, the legacy system backlog described by JLL is often less severe than in large enterprises. This can translate into a cleaner slate for adopting AI tools and embedding them into underwriting workflows, without the same level of technical debt.

As AI tools become more common in underwriting, the question is shifting from "Is the answer right?" to "Is the answer right the same way twice?" Altus Group put it plainly in August 2026: "It isn't enough for an AI tool to be right, he argued; it has to be right the same way twice. Ask an LLM the same question today and tomorrow, and there's no guarantee the answers will match, which is a real problem for CRE."

This reproducibility gap is not just theoretical. In CRE, underwriting decisions must be defensible and consistent—especially when presented to investment committees or lenders. As Altus Group notes, "You could go to a generic AI tool and ask questions about your portfolio. But could you take that output into an investment committee?" The answer, for most generic AI tools, is no. The risk is that outputs may vary from one run to the next, undermining trust and making it difficult to stand behind the analysis.

The challenge is compounded by the data environment. As Altus Group observed in September 2025, "Data serves as both the fuel and choke point for AI adoption and, with this in mind, firms must treat it as a governed asset rather than an afterthought. 'Data is where all advanced analytics projects go to die,' Blanco quoted, crystallizing a frustration frequently echoed across the industry." Without standardized, validated data, even the most sophisticated AI tools can produce inconsistent or misleading results.

The legacy system backlog described by JLL is largely an enterprise problem. For a mid-market Ontario acquisitions team, there may be fewer entrenched systems to unwind, making it easier to implement AI tools that deliver consistent, repeatable results. This relative agility can be an advantage, allowing these teams to move beyond experimentation and embed AI directly into their underwriting processes.

But the core question remains universal: Before your next investment committee meeting, can the underwriting you present be rerun tomorrow and land on the same answer? In a world where AI can generate professional-looking analyses in minutes, the standard for credibility is no longer just speed or even accuracy—it's repeatability and defensibility. As CRE firms continue to experiment with AI, those who prioritize these qualities will be best positioned to turn pilots into production and insights into action.

Key takeaways

• AI pilots in CRE have surged from 5% to 92% in three years, says JLL
• Only 5% of firms have achieved most AI program goals—most are still experimenting
• Legacy systems and inconsistent data are major barriers to scaling AI
• Repeatability, not just accuracy, is now the standard for AI-driven underwriting
• Mid-market teams may have an advantage due to less legacy tech debt