If you're a leader right now, AI probably feels like two things at once:

  1. I can't ignore this.
  2. I can't tell if we're doing it "right."
That tension shows up in SMBs and in large enterprises. The stakes are different, but the experience is oddly similar: lots of experimentation, lots of noise, and a nagging feeling that the ROI isn't matching the buzz.

That's why a recent Harvard Business Review study caught my attention. It starts with a reality check many orgs are living right now: AI use is widespread, yet adoption still stalls and performance gains plateau.

Too many orgs treat AI adoption like a purely technical rollout. It requires a more human approach, one with your employees at the center of it.

"We have usage" is not the same as "we have adoption"

One reason this hits is that it names a specific frustration: people are experimenting with tools but not integrating them deeply into how work gets done.

So, leaders see prompts flying, pilots launching, a few success stories, and then the gains flatten.

This is happening at scale, too. McKinsey's 2025 global survey found 88% of respondents said their organizations were regularly using AI in at least one business function, but only about one-third reported that their companies had begun scaling AI programs across the business.

That gap is the heart of the story: "AI is present" doesn't mean "AI is embedded."

HBR's argument: the stall is often a people-and-context problem

HBR's framing is that many companies assume the problem is execution (more training, more mandates, more governance). But the research they point to suggests initiatives often stall because employees experience AI as personally risky, shaped by their industry context and their sense of relevance/identity/job security.

That's a big perspective shift.

It suggests the barrier isn't always "we didn't educate people enough." Sometimes it's "people are protecting themselves."

And if that's true, you can end up with something that looks successful on paper (activity, compliance, even heavy use) while still falling short of durable workflow change.

The SMB lens: it can be psychological risk and operational risk

When I read this, I immediately thought about SMBs, and I don't mean that as a "small-business-only" problem. I mean SMBs are often the clearest place to see it, because everything is closer to the surface: Fewer layers of governance, faster tool sprawl, less margin for error, more direct accountability when something goes wrong, etc.

In SMB conversations, the risk often shows up as very practical questions:

"Is this secure?"

"Is this going to leak client data?"

"Will my team use this outside policy?"

"Is this complicated or expensive?"

"Where do we even begin?"

Those questions are not separate from the HBR point, they're often how "risk perception" manifests in real organizations.

If people think AI is a career risk, they may avoid it or use it quietly.

If leaders think AI is a business risk, they may stall the rollout or over-correct with policies that nobody understands.

Either way, the result can look the same: experimentation without integration.

What leaders may be underestimating

The practical takeaway I keep coming back to is HBR's emphasis that leaders who treat adoption as a psychological and contextual challenge, not just a technical rollout, are more likely to convert experimentation into sustained impact.

That doesn't mean "ignore governance." It means:

  • You'll never govern your way into trust.
  • You'll never train your way out of fear.
  • And you can't measure your way into adoption if your metrics only capture activity.

Why we're posting about this

We're sharing this because we've seen the stall up close, in our own organization and in others.

Not because AI is failing, but because adoption is messy, the landscape changes quickly, and most leaders are trying to make decisions with imperfect information.

Research like this is useful because it nudges the conversation away from "Which tool?" and toward the harder question:

What does AI feel like to the people we're asking to use it, and what risks (real or perceived) are they managing?

That's where adoption tends to either become durable... or stay stuck as "AI theater."

This post references Harvard Business Review's "Why AI Adoption Stalls, According to Industry Data" (February 17, 2026) and McKinsey's "The State of AI in 2025."