Why CRE Firms Keep Spending Millions on AI and Getting Little Back

Commercial real estate has entered the strange part of the AI cycle, where nearly everyone is experimenting, nearly everyone is paying attention, and many firms are spending real money on tools that, by most honest assessments, have not yet changed how the business actually runs.

The reason is not that AI lacks power and potential, but that much of the industry is trying to automate work it has never properly organized in the first place. It is pointing new technology at scattered data, inconsistent definitions, and workflows that have been held together for years by a combination of smart people, late nights and very complicated spreadsheets — and then finding itself surprised when the results are underwhelming (or wrong).

Deloitte’s 2026 Commercial Real Estate Outlook put a number to that disappointment: The share of CRE executives reporting a transformative impact from artificial intelligence fell from roughly 12 percent to about 1 percent in a single year, even as adoption continued to rise.

Firms are discovering the gap between using AI and actually becoming an AI-enabled company, and finding it considerably wider than the vendors suggested.

Using AI is straightforward enough: You buy licenses, give people access, run a pilot, and fairly quickly someone summarizes a lease faster than before, someone produces a draft report, someone builds a dashboard that looks impressive in a meeting. 

But becoming AI-enabled is something else entirely. It’s messier and far less glamorous, and it starts with the questions most organizations have spent years successfully avoiding: What exactly do we mean by this metric? Why does asset management describe this property one way and operations describe it another? Which number is the real one, where does the source of truth live, who owns it, and what happens when two teams are both right in their own context, but wrong when the information is combined?

AI does not make those questions disappear. Instead, it drags them into the open in ways that no amount of manual workarounds could have forced before, which is why the process of deploying it seriously can be, for many firms, unexpectedly clarifying.

That is also why so many implementations stall. A firm takes a general-purpose AI model, points it at documents, property data, budgets, leasing information and reporting packages, and expects intelligence to emerge from the exercise. But, if the underlying data is fragmented, the output will be fragmented, and if the process is unclear, the answer will be unclear. If the organization itself has not agreed on the rules, the model will not invent the discipline that was missing before it arrived. It will simply move faster through the confusion.

This is the part that tends to get lost in the marketing. There is a meaningful difference between generative output and investment-grade output, and, in commercial real estate, the distinction has financial consequences that are not abstract. A general AI tool can write a paragraph that sounds plausible, summarize a memo, and produce something that feels useful in a review meeting, but a wrong number in a rent roll, a mistaken assumption in a net operating income projection, or an invented data point in an asset management report is not merely embarrassing — it can be expensive.

That is where the return on AI starts to disappear. Whatever time a junior person saves by using the tool is often consumed later by a senior person checking the work, which means the bottleneck moves rather than goes away. In some cases, the firm has not automated the work at all but simply relocated the verification burden to people whose time costs more.

The industry conversation about AI tends to get stuck between two unsatisfying positions: those still in the marketing phase, convinced that any frontier model can be pointed at almost any business problem and made useful, and those who have become skeptics, having seen enough shallow pilots to conclude that the whole category is overhyped. Neither position is quite right, because the real question is not whether AI works but whether the organization has done the preparatory work required for AI to work inside that specific organization, with that specific data, in that specific operating context.

In commercial real estate, that means process first, data first, context first, and AI second. The firms that will get lasting value are not necessarily the ones that spend the most, but the ones that force the difficult internal reconciliation before expecting automation, the ones that clean up how data flows across systems, the ones that take the time to define the operating logic of the business explicitly.

Only then can AI do what people are hoping it will: surface patterns earlier, help asset managers understand controllable expenses, anticipate budget needs, compare vendor histories, monitor delinquencies, review leases, and support decisions with the kind of consistency that a manually assembled process, however talented the team, simply cannot sustain at scale.

AI can change commercial real estate in genuinely significant ways, taking people out of repetitive back-office work and giving them more time for relationships, strategy and judgment, which is where the industry has always created its real value. But it simply cannot compensate for organizational disorder.

Arunabh Dastidar is the co-founder and CEO of real estate investment platform Leni.