Here are two numbers that should not be able to coexist.
Roughly 40 percent of companies have an official AI subscription. Meanwhile, when MIT's Media Lab studied the state of AI in business, it found that at more than 90 percent of companies, employees were using personal AI tools for their daily work—often quietly, in the shadows, without approval, and often while the company's official AI initiative sits stalled in a conference room.
Read that again. The same employees labeled as "resistant to AI" are using AI before breakfast. They're just not using yours.
This is the single most important finding in technology right now, and almost nobody in leadership is talking about it, much less acting on it. Because it destroys the most comfortable explanation executives have for why their AI investment is underperforming: our people won't adopt it.
Your people adopt AI just fine. What they won't adopt is your rollout.
The evidence is brutal — and consistent.
MIT's researchers analyzed 300 public AI deployments, interviewed dozens of executives, and surveyed leaders across four industry conferences. Their conclusion: about 95 percent of corporate AI pilots produce no measurable impact on the P&L. Not disappointing impact. No measurable impact.
And before you assume MIT caught companies in a bad year, the pattern holds everywhere you look. PwC asked more than 4,400 CEOs across 95 countries what AI had done for their numbers; a majority — 56 percent — said it delivered zero cost or revenue improvement. A Writer poll found that nearly half of executives now privately describe enterprise AI adoption as a massive disappointment. S&P Global found that 42 percent of firms scrapped most of their AI initiatives in 2025, more than double the year before.
Here's the part that matters: in study after study, the failures are not explained by the technology. MIT said it plainly—the barriers are strategic, organizational, and cultural. The models work. The organizations don't.
The 52-point gap where your ROI went
The software adoption firm WalkMe surveyed thousands of executives and employees across 14 countries and found something remarkable: 61 percent of executives trust AI to help with complex decisions. Among their employees? Nine percent.
That is a 52-point trust chasm running straight through the middle of your organization. And it shows up in behavior: in the same research, more than half of workers admitted they had bypassed their company's AI tools in the previous 30 days and simply done the work manually. They didn't file a complaint. They didn't push back in the town hall. They just quietly went around the system you paid for—while, in many cases, using their own AI tools to do it in the shadows.
So here is the real picture. Leadership bought a capability. The organization built a workaround. And the gap between those two things—between what leadership bought and what the organization actually does with it—is where the money disappeared.
That gap has a name in my profession. We've been studying it for decades. It's called a change leadership failure. And I've watched it destroy value my entire career—long before anyone called it AI.
I've seen this movie before. I was in it.
In my twenties, I was an outside sales rep for a global Swiss pharmaceutical company, and I was handed the state of the art: a fourteen-pound laptop that could bruise your legs if you actually set it on your lap. It had 640 kilobytes of memory, 20 megabytes of storage, ran MS-DOS, and cost five thousand dollars.
A couple of years later I had barely learned my DOS commands when the company replaced it with something ten pounds lighter that came with a strange new device—a plastic "mouse" I was supposed to drag around my desk.
My colleagues were furious. They called the new machines unreliable, unnecessary, and a distraction from real work. But sitting in those meetings, I could see what was actually underneath the criticism: fear. The unspoken sentiment in every room was, I finally got comfortable with the old system, and now you're throwing me out of my comfort zone again?
I realized I had three options: resist, comply, or embrace. Resisting was career suicide. Complying—doing the bare minimum—was what most people chose. I chose to embrace it: I learned faster than required, then started helping my colleagues learn. That decision made me the district's trainer and eventually the first in my cohort promoted to management.
For years, I told that story as a lesson about attitude. But three decades of leading organizational change taught me the harder truth hiding inside it: my company left that choice entirely to chance. Nobody addressed the fear in those rooms. Nobody made the case for why. Nobody walked alongside the people who were struggling. The company got one embracer, a roomful of compliers, and a couple of resisters—purely by luck of individual temperament.
That is exactly what's happening with AI right now, at a global scale, with billions of dollars on the line. Companies are buying the technology and leaving resist-comply-embrace to chance. And chance, statistically, delivers compliance: the bare minimum, the quiet workaround, the pilot that dies in a slide deck.
We've known the fix for decades.
This is the maddening part. McKinsey studied companies running major change initiatives and measured ROI against the quality of their organizational change management. With excellent change management plans, initiatives returned 43 percent. With poor or absent change management: negative 65 percent. A 108-point swing—and none of it was explained by the technology being deployed.
The successful companies shared three traits:
· Leaders at every level were genuinely involved.
· Responsibilities were clear, and
· People throughout the organization understood and accepted why the change was happening.
That's it. That's the secret that's been sitting in the research for decades while companies burn their AI budgets solving the wrong problem.
Because make no mistake—most AI implementations today have excellent project management and nonexistent change leadership. There's a timeline, a budget, a vendor, and a training module. There is nobody whose job is to close the 52-point trust gap. Nobody is asking employees what they're afraid of. Nobody is redesigning the workflow so the AI serves the people doing the work, instead of being bolted on top of them.
Close the gap or keep funding it.
If you're leading a company between one and twenty million in revenue, you don't have a Fortune 500 cushion to absorb a failed initiative. You also have an advantage they don't: you're close enough to your people to actually lead this change instead of just announcing it.
Start where we start with every client—with diagnosis, not tools. What business problem are you actually solving? How will we measure success? Where is the real bottleneck in your system? Can your people articulate why this change is happening—and do they believe you when you say it's not about replacing them? What are they already doing with AI on their own that your official rollout should learn from?
That last question is the one almost nobody asks. Your shadow AI users aren't a governance problem to stamp out. They're a free, running pilot of what your people actually find useful—evidence, sitting in plain sight, of where the gap between what you bought and what they need is widest.
Your employees have already voted. They believe AI is useful. What they don't yet believe in is the way it's being handed to them. Closing that gap isn't a technology project. It's a leadership act—diagnosing honestly, operating precisely, and staying with your people until the new way of working actually heals into place.
That's the work. It's harder than buying software. It's also the only version of this that pays.
Douglas Lundrigan has spent 30 years helping organizations lead change—through mergers, technology shifts, and now AI transformation. He is a co-founder of Pulse AI. If your AI investment isn't producing what you expected, book a 20-minute AI Readiness Pulse Check: no pitch, just a direct, honest diagnosis of where your effort is stuck.
