AI Adoption Barriers in Enterprise CRE Firms

JLL's 2025 Global Real Estate Technology Survey found 88% of CRE investors and owners have launched AI pilots. Only 5% report hitting most of their program goals. That gap has nothing to do with enthusiasm or budget, since 87% of firms are increasing AI spend right now. Across the available data, the pattern holds up: the reasons are structural, they live in process rather than culture, and they're fixable if a firm is honest about where it actually stands.

A separate survey by Keyway and The Appraisal sharpens the picture. Forty-five percent are running pilots. Only 9% have reached enterprise-wide deployment. Trials get launched constantly; what almost never follows is the infrastructure needed to turn a trial into something a firm actually relies on, and that's what the rest of this piece tries to unpack.

How chasing five use cases at once compounds the execution problem

JLL found that firms pursuing AI are chasing an average of five use cases at once. That gives a program five separate ways to fail before it produces anything worth showing a managing director.

Running five pilots side by side feels productive. It rarely is. No single workflow gets the depth of integration needed to prove a real return, so underwriting intake gets a shallow AI layer, comp analysis gets another, document generation gets a third, and none of them go deep enough to change how an analyst spends a Tuesday. The pattern repeats: underwriting intake, loan sizing, and market comp automation pursued in parallel, and the result is multiple mediocre pilots instead of one strong one. Each half-finished pilot becomes ammunition for the skeptics in the room rather than proof for anyone hoping the thing works.

The deeper mistake is treating AI as a basket of experiments rather than a capability built in sequence, one workflow proven before the next even gets funded. The disciplined shops pick a single high-volume, measurable process and drive it to repeatable results before touching a second use case. It looks slower in month one. By month twelve, it's the only version still standing.

Underneath that scope problem sits a second, less visible one: the data the workflow actually runs on.

Why data infrastructure — not organizational willingness — is now the primary constraint

Gartner projects that through 2026, organizations will abandon 60% of AI projects that lack AI-ready data. Data infrastructure functions as a prerequisite rather than a parallel workstream run alongside AI deployment. Firms treating it as an afterthought are choosing to fail early instead of late, which at least has the mercy of being cheap.

CRE has a worse version of this problem than most industries realize. Fifty-four percent of CRE teams cite compatibility with legacy infrastructure as their top barrier to AI progress, and 81% report at least three existing systems that already fail to produce expected results, before AI enters the picture at all. Anyone who has assembled a deal package knows why. One file comes as a Yardi export, another arrives as a scanned PDF a broker typed up by hand, a third is a bespoke Excel model built by an analyst who left the firm two years ago and never documented a line of it. No two documents look the same, and that inconsistency isn't a minor inconvenience: a system configured for one document shape fails on the next, often silently, with nothing flagging that it happened.

Only 8% of surveyed CRE professionals described their data infrastructure as fully ready for AI deployment. AI-ready, in this business, means normalized property financials, consistent field mapping across asset classes, and document provenance you can trace back to the source PDF or export. Firms that standardized their data years ago, often for reasons that had nothing to do with AI, are the ones positioned to move fast now. Everyone else is doing that work retroactively, under deadline pressure, which is the hard way to learn it.

Workable data doesn't mean anyone trusts what comes out the other end of it, though.

The trust gap that keeps AI output out of investment committee packages

Trust in AI outputs ranks among the three biggest barriers keeping pilots from becoming standard practice, alongside data readiness and system integration. Most industries would file that under usability. CRE finance can't. Investment committees and lenders built entire careers on human-sourced analysis, and they sign multi-million-dollar decisions on the strength of it. Asking them to trust a black box asks them to devalue what their own judgment is worth.

The trust test in institutional shops doesn't bend: if an AI-extracted figure can't be traced back to a source document, it doesn't make it into the credit memo. No exceptions. Explainability functions as the gatekeeping condition that decides whether AI gets anywhere near a regulated or committee-reviewed decision in the first place, well beyond a feature that makes a tool nicer to use.

Generic AI tools fail this test by design. They hand back an answer with no chain of custody behind it, which forces analysts to re-verify everything by hand anyway, quietly erasing whatever time the tool was supposed to save. Trust gets built at the institutional level through audit trails tying every number to its source document, through consistency across dozens of deals rather than one lucky accurate output, and through answers built to survive a skeptical credit officer asking where a number actually came from.

Trust also collapses for a simpler reason that has nothing to do with the software: nobody taught the people using it how.

The internal expertise gap and what "preparing for AI" actually means in practice

JLL's 2025 survey found that only 33% of the CRE workforce feels adequately trained on AI. Sixty-two percent of organizations report preparing for AI in some form, but more than half of those firms give staff no actual training. Preparation, too often, amounts to a memo and a software license rather than any real investment in people.

A 2025 DataSociety AI readiness survey found that 65% of organizational leaders didn't know when or where to apply AI, and 52% lacked even a foundational understanding of how it works. Ninety percent of institutional investors have built, or are building, AI-focused teams, yet 93% still cite significant adoption barriers. Standing up a dedicated AI team doesn't fix a workforce-wide skills gap; it just concentrates the knowledge in a conference room most analysts never walk into.

An analyst who can underwrite a deal cold cannot necessarily prompt an AI tool well, validate what it hands back, or catch the moment it quietly gets something wrong. Those are different muscles, earned differently, and firms assuming underwriting skill transfers automatically are setting their own pilots up to disappoint everyone involved. Real training looks like analysts running the tool on live deals under supervision, with a feedback loop catching errors before they reach a committee. A one-hour AI literacy seminar that checks a compliance box changes nothing.

Build that expertise, and there's still a harder problem waiting underneath it.

Why generalist AI tools and implementation partners fail CRE deal workflows

A leading cause of CRE software deployment failure is picking an implementation partner with no real estate expertise, one that configures a system around generic best practice instead of the way deals actually move through that specific shop. The result is predictable: low adoption, analysts quietly building workarounds that recreate the manual process the tool was supposed to kill, and eventual abandonment that confirms every skeptical executive's suspicion that AI ROI was overhyped from the start.

Generic AI tools bolted onto CRE workflows hit the same wall from a different angle. They can't handle asset-class-specific policy, can't normalize the particular chaos of CRE document formats, and can't enforce the deal logic separating a multifamily underwrite from an industrial one. A generic extraction tool pulls text off a page. A purpose-built CRE agent knows what a rent roll is supposed to look like, knows what to flag when NOI doesn't reconcile against a trailing twelve, and knows how to size a loan against one lender's actual policy instead of a generic template pulled from nowhere in particular.

Parsing text and understanding a deal are not the same skill, and that gap is most of the game. Platforms built around real-time CRE data and workflow-specific agents, for lender search, deal document production, loan sizing, solve the integration problem at the foundation instead of papering over it. Configuring a generalist tool to approximate CRE logic just moves the failure point downstream, delaying the invoice rather than removing it.

This mismatch between tool and workflow explains a good chunk of the ROI numbers that have pushed so many firms to walk away from AI entirely.

What the ROI failure rate actually reveals about how firms are measuring AI

Recent analysis found that 95% of generative AI pilots failed to deliver measurable financial return. IBM put the share of initiatives hitting expected ROI at just 25%. S&P Global found that 42% of companies abandoned most of their AI projects in 2025, more than double the 17% that abandoned projects in 2024. That abandonment rate isn't leveling off, it's accelerating.

Those numbers say less about AI's ceiling than about how badly firms are measuring it. Grading a pilot against broad financial return inside a short pilot window is the wrong test, full stop. AI's value in CRE compounds through workflow throughput, deal velocity, and analyst capacity over time, not as a line-item cost reduction in quarter one. Firms grading a pilot on that basis will call it a failure almost by construction, before it ever had a real chance to prove itself.

Gartner's findings back this up: organizations with high AI maturity were more than twice as likely as low-maturity organizations to keep initiatives running past three years. Persistence and accumulation are the actual mechanism by which ROI shows up, not the pilot quarter, and firms that quit at month six never got close to finding out. Mature CRE AI programs measure time from loan application to credit committee, the number of deals an analyst clears per week, and how AI-generated credit memos stack up against manually produced ones on accuracy and completeness. The firms winning on ROI run fewer pilots than everyone else, go deeper on each one, and tie every metric to a workflow output instead of a company-wide number dreamed up in a boardroom.

A structural framework for moving from pilot to enterprise scale in CRE

The path from pilot to scale follows from the barriers above, and the order matters. Skip a step and the next one collapses on top of it.

Start with data, before any vendor gets a phone call. Audit existing systems for normalization gaps, legacy compatibility failures, and provenance holes first, because the 8% of firms with fully AI-ready data didn't get there by deploying AI and hoping the data caught up on its own. They did the boring part first, the part that never shows up in a case study. From there, narrow the scope before scaling anything. Pick one workflow where volume is high, outputs are measurable, and an error is recoverable rather than catastrophic: intake processing, comp analysis, initial loan sizing. Prove repeatable results there before adding a second use case to the pile.

Judge tools on CRE specificity rather than general capability or a slick demo that impresses a committee for twenty minutes and solves nothing afterward. The right question isn't whether a tool can extract data from a document, it's whether the tool understands what to do with a rent roll, an operating statement that has drifted from its T-12, or a borrower entity with three layers of guarantors buried inside it. Build trust through transparency by requiring source-linked outputs from anything touching a deal workflow, since anything that can't be traced gets re-verified by hand anyway, which kills the time savings the tool was bought for. And define the ROI metrics on day one, not after the fact: deal velocity, analyst throughput, time-to-credit-decision, each judged on its own terms rather than folded into some aggregate financial number due by year-end.

Get the sequence right, and failures shrink as the program grows instead of stacking on each other. A unified platform combining CRE-specific data, purpose-built agents, and deal document workflows removes the integration layer as a failure point by design, rather than by accident. Stitching together point solutions that each need their own upkeep just relocates the same structural problems this piece has walked through, dressed up as progress.