Almost every institutional real estate investor now says AI is part of how deals get made. Far fewer say it has made them faster. In a vendor survey released this summer, 97% of respondents said AI is built into their investment process, yet only about half said it saves them time once they count the work of checking its answers. Four in ten said work with AI in it now takes longer than doing it by hand.
That gap has a name in the industry now: the verification tax. Every AI output has to be checked before anyone trusts it with a decision. In commercial real estate the checking is slow, because the data underneath is scattered across systems, spreadsheets and inboxes. For most mid-market firms the answer is not another AI tool. It is the plumbing the tools draw from.
The Number Behind the Headline
The figures come from Dealpath’s 2026 State of AI in CRE Investing survey, released July 8, 2026, and later summarized by IREI. Dealpath sells deal management software, so read this as a vendor survey with a point of view. It is also small. The sample was a little over 100 investment and technology professionals at REITs, private equity real estate firms, pension funds, insurers and banks in the US and Canada, at firms managing from $500 million to more than $40 billion. Every respondent already uses AI in their role. The 97% describes AI users, not the whole market.
Even with those limits, the pattern inside the Dealpath numbers deserves a careful look. According to the release, 51% of respondents say AI saves them time once verification is counted, while 41% say AI-involved work takes longer than doing it manually. When asked why AI falls short, 43% pointed first to fragmented data, ahead of made-up answers or the limits of the models themselves. The release adds two more figures that sit uneasily together: 90% say data quality or fragmentation has limited AI’s impact at their firm, and 83% rate their own data as mostly or fully ready for AI.
Read those last two side by side. Nearly everyone says data problems are holding AI back, and most also say their own data is ready. Both cannot be fully true at the same firm. The more likely reading is that people judge readiness by whether the data exists, not by whether a machine can find it, trust it and reconcile it. Dealpath’s CEO, Mike Sroka, put it plainly in the release: “the data problem persists.”
Nearly everyone says data problems are holding AI back, and most also say their own data is ready. Both cannot be fully true at the same firm.
Adoption Is Wide, Not Deep
Larger studies point the same way. Deloitte’s 2027 Commercial Real Estate Outlook, based on a survey of 950 C-level executives in June and July 2026, found that 92% of firms are still researching or piloting AI. Only 8% said they had integrated it. IREI’s coverage of that report adds that more than 90% plan to raise data and technology spending next year. So the money is coming. The open question is where it lands.
IREI has also raised a sharper concern: the growing pile of AI tools inside CRE firms may be slowing adoption instead of speeding it up. That claim fits what the surveys describe. The picture is a wide, thin layer. Nearly everyone has a chat assistant. Few have AI wired into the systems where leases, rent rolls, loan files and deal pipelines actually live.
Wide and thin is not a failure. It is a stage. Most technologies spread this way, first as personal tools and only later as part of the operating core. The risk is mistaking the first stage for the finish line, and then buying more of the same thing in the hope that depth will follow from breadth. It rarely does.
Where the Verification Tax Comes From
Picture what an analyst does when an AI tool drafts a deal summary. This is an illustration, not a case, but it will look familiar. The rent roll came from one system. The operating statement came from a spreadsheet a property manager emailed last month. The comps came from a data vendor with its own field names. The loan terms live in a scanned PDF. The AI may have pulled from some or all of those, and the analyst has no easy way to see which. So the analyst opens each source and checks. That hour is the tax.
Now multiply it. A firm with a general chat assistant, a separate underwriting helper, a lease abstraction tool and an AI add-on to its broker CRM has four places where answers come from. Each has its own habits, its own gaps and its own way of being wrong. Each tool may be perfectly sound alone. Together they create more reconciling, not less, because every new source of answers is also a new source of doubt.
The tax falls hardest on your best people. The senior analyst who knows which spreadsheet to trust becomes the human index for the whole firm. That knowledge is valuable, and it does not scale. When that person is on vacation, the checking slows down or, worse, stops.
The tax falls hardest on your best people. The senior analyst who knows which spreadsheet to trust becomes the human index for the whole firm.
This is the same lesson we laid out in Getting Started with AI · A Guide for CRE Firms: data has to be treated as an asset before AI can pay off. What has changed since then is scale. The firms that skipped that step did not stay out of AI. They bought it anyway, and now they pay the tax on every output.
Why the Tax Costs More in a Tight Market
Timing makes the tax more expensive. KBRA’s September 2026 CMBS loan performance report put the overall delinquency rate for the private-label CMBS loans it tracks at 7.7%, or $25.8 billion, with the office distress rate at 17.6%. Those numbers describe one slice of the debt market, not every property, but the direction is clear enough for planning purposes.
In a market like that, lenders, asset managers and brokers need to move with care and with speed on workouts, refinancing and repositioning. Speed built on bad data is a risk. Slowness from checking everything twice is a cost. You do not get to choose neither; you only get to choose which one you carry. Clean, connected data is how a firm gets the speed without the risk, and it is the one investment that makes every later AI purchase cheaper to trust.
The CEO Move: Fund the Plumbing Before the Next Tool
The fix is not glamorous, and it is not a product purchase. It is a handful of decisions that only leadership can make, because each one crosses a departmental line that no single team can cross alone.
Start by counting the tools. Ask for a list of every AI tool and AI feature in use across the firm, paid or free, with the names of the people who actually use each one. Most firms find more than they expected, often because AI arrived quietly inside software they already owned. Choose the few that matter to your core work and retire the rest. Fewer sources of answers means fewer things to check.
Then name the source of truth for each core record. Leases, rent rolls, property operating data, loan terms, comps and contacts each need one home system that everyone agrees wins when the numbers conflict. This sounds administrative. It is not. It is a judgment about which team owns which fact, and it will surface disagreements that have been politely ignored for years. That is the point.
Next, connect before you add. Put budget into the links between those systems and into a shared data layer before approving another AI tool. An AI tool that reads from one trusted place needs far less checking than one that reads from five uncertain ones. If a vendor pitches a new assistant, the first question is not what it can do. It is where it will get its facts.
Measure the tax while you work. Ask a few analysts to track, for two weeks, how long they spend checking AI output against source files. That figure becomes your own baseline, which is worth more than any industry survey because it describes your desks and your data. It also tells you where the plumbing hurts most, so the first dollars go to the worst leak.
Finally, give one person the job. Data ownership spread across IT, finance and the deal teams usually means no one owns it. A named owner with the authority to settle conflicts moves faster than a committee, and a CEO who backs that person in public moves faster still.
None of this means pausing AI. It means pointing the next dollar at the foundation instead of the next feature. Our regional CRE brokerage case study shows what that sequence can look like when a firm builds the data layer first and lets the AI arrive on top of it.
What a Mid-Market Firm Should Expect
If you run a firm in the middle of the market, you face a particular version of this problem. You have more systems than a small shop and fewer people than a large institution. You likely do not have a chief data officer, and your IT lead is already stretched across security, vendors and help desk. The data work tends to fall to whoever cares most, and that is rarely a plan.
The good news is that the middle of the market can move faster than the giants. Your list of core records is shorter. Your vendor contracts are fewer. A decision made in your office on a Monday can show up in your systems within a quarter. As an illustration, a firm with a few hundred assets might find that rent rolls and loan terms account for most of its checking time, fix those two links first, and see the tax fall long before the full data layer is done. The exact result will vary. The sequence holds.
Expect the first gains to be quiet. Fewer emails asking which number is right. Shorter review cycles on investment memos. Analysts who trust the summary enough to spend their time on the judgment instead of the arithmetic. Those are the signs that the tax is falling, and they usually come before any headline metric moves.
Expect the first gains to be quiet. Fewer emails asking which number is right. Shorter review cycles on investment memos.
What This Will Not Do
Fixing the data will not make AI perfect. Models still make mistakes, and some checking will always be needed for decisions with real money on the line. That checking should be a deliberate step, not a scramble through five systems.
Consolidating tools also has costs. There are contracts to unwind, people who liked the tool being retired, and a few months where things feel slower before they feel faster. Name those costs out loud so they do not become reasons to stop halfway.
And the surveys cited here measure what people say, not what was timed on their desks. Dealpath’s survey in particular is small, drawn only from people who already use AI, and published by a vendor. Treat the numbers as a strong signal, not a precise benchmark. Your own two-week measurement will tell you more.
Three Questions for Monday
If you lead a CRE firm, ask your team three things this week. How many AI tools are we paying for, and which ones do people actually use? When two systems disagree on a rent roll or a loan balance, which one wins? And how many hours a week do our analysts spend checking AI output against the source?
If those answers come back fuzzy, the next AI purchase will likely add to the tax rather than cut it. An independent Innovation Vista assessment can map your data, your tools and your verification load, and show you where the first fix belongs.
