KEY TAKEAWAYS
- 92% of CRE teams are piloting AI or planning to, yet only a small share have achieved all their AI program goals, and most describe themselves as unprepared.
- Three failure modes explain most stalled AI programs: teams don’t adopt it, teams don’t know what to do with it, and teams fall back into old habits under pressure.
- The fix is architectural, not motivational: workflows that fit inside existing systems, a defined catalog instead of scattered pilots, and a named owner accountable for adoption.
- Sophia CRE’s own history, built internally first, then coached to other firms, then productized, reflects this exact adoption problem being solved in practice.
IN THIS ARTICLE
Almost every CRE firm is piloting AI in some form right now. Almost none of them would say the rollout actually succeeded. That gap, between widespread adoption of AI tools and a much smaller number of firms who’ve actually changed how they operate because of it, is the single biggest missed opportunity in commercial real estate right now.
It’s tempting to blame technology. It’s rarely the right diagnosis. The models are capable enough today to handle real underwriting, real lease abstraction, real reporting work. What fails almost every time is the implementation, not the AI itself. There are three specific failure modes behind most stalled AI programs in CRE, and all three are fixable, but only if a firm recognizes which one it’s actually dealing with.
Failure Mode 1: Teams Don’t Adopt It
A firm buys a tool, issues logins, sends a training email, and usage quietly dies within a few weeks. This is the most common failure mode and the easiest to misdiagnose as a training problem, when it’s almost always a workflow problem.
People don’t resist new tools out of stubbornness. They resist tools that sit outside the way they already work, that require an extra login, an extra tab, an extra step that has to be remembered on top of everything else already on their plate. If using the AI tool takes more total effort than just doing the task the old way, the old way wins every time, regardless of how much time the AI tool would theoretically save.
This is compounded in CRE specifically because the hardest domains for AI adoption tend to be property operations, management, and marketing, the exact functions where the day-to-day work is already fragmented across a dozen small tasks and a dozen different systems. Adding one more disconnected tool into that mix doesn’t reduce fragmentation. It adds to it.
The fix isn’t a better onboarding deck. It’s making the AI capability show up inside the workflow people already use, so using it is the path of least resistance rather than an extra chore layered on top of an already full day.
There’s also a trust dimension that gets overlooked. A tool that occasionally produces a confidently wrong answer, with no clear way to verify it, teaches people to stop trusting it fast, even if it’s right the other nine times out of ten. Adoption doesn’t just require convenience. It requires a visible, reliable review step, so the team knows exactly how much to trust an output and doesn’t have to silently wonder every time.
Failure Mode 2: Teams Don’t Know What to Do With It Once They Have It
This failure mode looks different from the first one but comes from the same root cause. The tool gets used, occasionally, for whatever task happens to come to mind that day, but never becomes part of a defined process. Nobody has decided which tasks it handles, what a good output actually looks like, or how the AI’s work fits into the broader workflow around it. Without that definition, the tool becomes a novelty rather than infrastructure. Someone remembers to try it when they think of it, and forgets about it the rest of the time.
This is what the industry, somewhat accurately, calls pilot purgatory. A firm ends up running several AI pilots simultaneously, none of which ever graduate into something the whole team relies on every day. Each pilot solves a narrow, specific problem for a specific person, and none of them get industrialized into a repeatable process the rest of the organization can depend on.
The pattern shows up almost identically across firms: enthusiasm at the start, a flurry of experimentation, and then a slow fade as the pilots that worked never get formalized and the pilots that didn’t work never get diagnosed. Six months later, the firm has spent real time and budget and can point to almost nothing that changed about how work actually gets done.
The fix here is architectural, not motivational. Rather than scattering effort across five separate pilots that each solve one narrow problem, a firm needs one catalog of defined, documented workflows, what a mature AI operating model calls Skills, each with a clear scope, a clear owner, and a clear definition of what a correct output looks like. Every new AI use case gets built to run on that same catalog instead of becoming its own disconnected experiment. That’s the difference between a firm that has five pilots and a firm that has one platform.
It also changes how a firm evaluates whether AI adoption is working at all. “Are people using the tool” is the wrong question, because people can technically use a tool occasionally and still get almost no organizational benefit from it. The better question is “how many of our defined workflows now run through this catalog, and how many still run the old manual way.” That’s a number a firm can actually track quarter over quarter, and it’s the number that correlates with real time saved, not just tool logins.
Failure Mode 3: Teams Fall Back Into Old Habits
Even when a workflow gets built and people know how to use it, the path of least resistance always wins under pressure. The moment there’s a tight deadline, an unfamiliar edge case, or a bad day, people revert to email, spreadsheets, and manual review, because that’s the deeply grooved habit, and the new process hasn’t had time to become the deeply grooved habit yet.
This isn’t a character flaw in the team. It’s what happens to any new process that doesn’t have a structural reason to stick. Without a named owner accountable for the AI program’s actual usage, and without a real review discipline that makes the new way the expected way rather than the optional way, backslide isn’t a risk, it’s the default outcome.
What actually prevents this isn’t a tool subscription or a one-time training session. It’s an operating discipline: someone specifically accountable for AI adoption across the firm, usually a Head of AI Operations or an equivalent role reporting to the CEO or COO, combined with a real human-in-the-loop review channel where every consequential AI-assisted action gets logged and approved as a matter of routine, not as an exception. When the review step is built into the workflow itself rather than bolted on as an afterthought, falling back to the old way stops being the quiet default and starts being a visible deviation someone would actually notice.
This is also where leadership involvement matters more than most CEOs expect it to. A program with no executive owner tends to drift back to old habits the first time a deadline gets tight, because nobody with real authority is watching whether it stuck. A program with a named, accountable owner and a standing review cadence tends to survive exactly the kind of pressure that would otherwise cause backslide, because the deviation gets caught and corrected before it becomes the new normal.
Why This Matters More in CRE Specifically
These three failure modes aren’t unique to real estate. But CRE has a particular vulnerability to all three because the work itself is already so fragmented across systems that don’t talk to each other. Acquisitions runs on one set of tools, asset management on another, accounting on a third, investor relations on a fourth. A generic AI tool dropped into any one of those silos in isolation almost never survives contact with how fragmented the rest of the operation already is.
That’s also why the fix has to be structural rather than tool-by-tool. Adding a smarter point solution to an already fragmented stack doesn’t reduce the fragmentation, it adds one more disconnected piece to manage. What actually works is a single orchestration layer that sits across every stage of the deal lifecycle, acquisition, operations, leasing, reporting, and disposition, with defined workflows and a consistent governance model running underneath all of it.
What Successful Implementation Actually Looks Like
The firms that get past all three failure modes tend to share the same pattern, regardless of asset class or size. They don’t start with a company-wide rollout. They start with a small number of well-defined workflows, running inside the tools the team already uses, with a named owner accountable for whether people are actually using them, and a real review process that makes correct usage the path of least resistance rather than an extra step people have to remember.
Sophia CRE’s own history reflects this pattern directly, because the platform wasn’t built as software first and sold second. It started as an AI-native operating system built internally for one firm’s own portfolio, then became a coaching practice helping other CRE firms work through exactly these adoption problems, then became a structured workforce training program, before finally becoming the productized platform it is today. Every stage of that evolution existed specifically because implementation, not the underlying AI capability, was the actual bottleneck the whole way through.
That’s the practical argument for treating AI adoption as an operating model change with a technology component, rather than a technology purchase with an adoption afterthought. The firms that get this right don’t ask “which AI tool should we buy.” They ask “which of our workflows are we going to redesign first, who owns making sure it sticks, and how will we know in thirty days whether it actually did.”
If your last AI pilot quietly faded out and nobody could quite say why, it’s worth revisiting which of these three failure modes it actually hit. The diagnosis usually points straight at the fix.
A Short Diagnostic for Your Own Team
Three questions tend to reveal which failure mode a stalled AI program actually ran into, rather than treating the whole thing as one vague disappointment:
- Did people stop using it because it was more work than doing the task manually, or because it was never clear what it was actually supposed to be used for? That distinguishes Failure Mode 1 from Failure Mode 2.
- Was there ever a written definition of what the workflow covered and what a correct output looked like, or did it live entirely in one person’s head? Missing documentation is the clearest sign of Failure Mode 2.
- Was there a named person accountable for whether the team kept using it three months in, or did it quietly become optional the first time things got busy? That’s the tell for Failure Mode 3.
Most firms discover the answer is some combination of all three, which is exactly why a single point-tool fix rarely holds. It takes a workflow that fits inside existing systems, a defined catalog instead of a scattered pile of one-off pilots, and a named owner who keeps the discipline in place after the initial excitement fades.
Frequently Asked Questions
Is this a technology problem or a people problem?
Neither, on its own. It’s a workflow design and accountability problem. The model is usually capable enough. What’s missing is a defined process, a place for it to live inside existing tools, and someone accountable for whether it’s still being used in three months.
What’s the fastest way to tell which failure mode we’re in?
Ask whether people stopped using a tool because it was extra work, because nobody defined what it was for, or because nothing held the new habit in place once things got busy. The answer usually points straight at one of the three modes.
Does fixing this require a company-wide rollout?
No. Firms that succeed typically start with a small number of well-defined workflows inside tools the team already uses, with a named owner, then expand from there rather than launching everything at once.
Who should own AI adoption inside a middle-market firm?
Usually a Head of AI Operations or equivalent role reporting to the CEO or COO, someone specifically accountable for whether defined workflows are actually being used, not just whether licenses were purchased.
Stop the Backslide. Build Real Adoption.
Sophia CRE helps you build defined workflows inside the tools your team already uses, with a named human reviewer on every step so adoption actually sticks. Book a conversation to see how it works.
Orchestration layer. Defined workflows. Named human reviewers.