The AI Adoption Problem CEOs Are Misdiagnosing

I interviewed 15 C-suite leaders across various industries, and not a single one is stuck on whether AI matters to what they’re doing. They are, however, quite stuck.

My interviews were with CEOs and near CEOs, individuals who run all or big parts of organizations from $5 million trade associations to $400 million consumer brands. I’m not a professional researcher, but I’m definitely an asker. I posed the same five questions to each person, curious if I might learn something about AI adoption worth sharing.

Almost nobody needs to be convinced that AI is important. Yet there is a real gap between knowing it’s important and knowing what to do next, at least beyond authorizing people to use their built-in Copilot for meeting notes.

The more interesting question is why.

Where Businesses Are Actually Stuck

Academic models are not my natural way of thinking. I’m more an artist in businessperson’s clothes. But sometimes a framework can reveal what otherwise might be missed.

There’s a well-known model called ADKAR that maps what people need in order to move through organizational change: Awareness, Desire, Knowledge, Ability and Reinforcement. The stages build on one another. Train someone who hasn’t emotionally committed, for instance, and the return on that training will be limited at best.

When I compared the notes from my fifteen conversations to ADKAR, one thing was clear: almost nobody was really stuck at Awareness. They all know AI matters. I detected little to no denial about that.

Where they were stuck was further downstream, particularly around Desire and Knowledge.

Organizational desire is different from personal desire. The CEO of a $28 million business in health and wellness explained that the fear of failure baked into the organization’s culture is perhaps particularly activated by AI. The CEO of a nonprofit told me her team can’t get serious about AI until they’ve wrestled to ground societal implications such as the risk of bias in predictive AI and the environmental impact of data centers.

These aren’t primarily technology problems. They are innovation problems, commitment problems and, in some cases, philosophical ones.

But across the interviews, something more specific emerged. Many companies may be misdiagnosing their AI problem. They assume they lack capability when, in fact, they haven’t yet created the conditions that allow capability to turn into coordinated action.

Four Sticking Points

Across my conversations, four sticking points emerged.

Permission paralysis

In almost every conversation, leaders told me one of two things. Either their early enthusiasts, the people using generative AI right now, are reluctant to say so, fearing they’ll be seen as cutting corners. Or their people who are not using AI yet are reluctant to say so, fearing they’ll be seen as behind or anti-progress.

The CEO of a trade association for outdoor goods and equipment told me both camps in his organization are staying quiet.

That’s not an awareness gap. It’s a permission gap.

And the consequence is larger than hesitation. If both adopters and non-adopters have reasons to stay quiet, leaders may not have an accurate picture of the organization’s real AI maturity. They can’t see clearly where capability exists, where resistance exists or what employees actually need.

Before leaders can accelerate adoption, they may first need to make the current state visible.

Tyranny of the urgent

A president of North American operations of a consumer-packaged goods company named this directly. AI stays perpetually in the someday pile because today’s fires arrest available attention.

Leaders care, but nothing forces AI strategy onto this week’s calendar.

The problem isn’t necessarily conviction. Organizations naturally privilege issues with immediate accountability over opportunities whose payoff is uncertain and whose ownership is unclear. AI can therefore be strategically important and operationally optional at the same time.

That is a dangerous combination.

Fragmented learning

One business-unit leader described her organization as “speaking different languages” on AI. Marketing uses one tool for briefs, creative uses another for imagery, operations is experimenting with a third.

Experimentation itself isn’t the problem. The problem is that the learning remains local.

One function discovers something useful, but the insight doesn’t necessarily travel. Another repeats the same experiment elsewhere. A third reaches a different conclusion because it uses a different tool, process or standard of success.

The organization can therefore generate considerable AI activity without accumulating much institutional knowledge.

The strategic issue is not tool fragmentation so much as fragmented learning.

Ownership vacuum

Several leaders I spoke to genuinely could not name who in their organization was accountable for driving AI forward.

Is it IT’s job? Operations? Marketing?

This is hardly unique to AI, but AI makes the problem unusually visible. When an opportunity touches nearly every function, it can easily belong to none of them.

And when no one owns the next decision, broad enthusiasm becomes a substitute for progress.

The Knowledge Problem

Knowledge was the other major stall zone.

These leaders believe AI matters. Many genuinely want to move. But they don’t know where to place the first meaningful bet. They don’t know if they are ahead of competitors or behind. They don’t know which investment option would be most strategic or where meaningful business value might actually reside.

Analysis paralysis sets in.

The CEO and cofounder of a company that owns and operates more than 100 urgent care centers told me their team rates itself 0–1 on AI expertise. They invested in one enterprise-level AI tool rather quickly, and it’s working. But without embedded expertise, what comes next is unclear.

“I’d love someone to show up having scoured our industry, with 20 use cases and real ROI data,” he explained. “We don’t have the bandwidth to do that homework ourselves.”

The CRO of a fast-growing SaaS company told me they’re idling at a crossroads: invest in AI to help their overtaxed developers, or build features that give the platform’s end users more visible AI capability.

Another CEO described AI as something his entire leadership group knows they should chase and then candidly admitted that, strategically, that is as far as they’ve gotten.

This is where the distinction between capability and readiness matters.

A company can have access to excellent tools and willing employees and still not be ready to make good AI investments. Readiness requires something different: a way to compare opportunities, establish ownership, define acceptable experimentation and learn quickly enough to make the next decision better than the last.

More tools do not solve that problem.

What the Unstuck Did Differently

One head of IT for a global consumer products business generated early uptake quickly when the team rolled out a small generative-AI tool, with more than 55 percent of employees at least accessing it.

Access alone does not prove transformation. But the rollout gave the company something many of the more stalled organizations lacked: a starting point from which to learn.

She explained that the company did three things to get going.

First, they planted a flag. Leadership declared to its workforce that AI mattered, that it was sanctioned and that its use was encouraged. Behind that was a clear belief that if more of their 800-plus people used generative AI more strategically and more often, the business could gain an edge.

Simply saying something official sounds obvious, but many of the stuck organizations had made no explicit declaration at all.

Second, they created guideposts. A clear policy distinguished between acceptable uses and off-limits territory, giving people confidence to experiment without fearing they were stepping outside the lines.

Third, they started small. Rather than launching a sweeping transformation initiative, they identified a few use cases, ran pilots and learned from them before making larger commitments.

The significance of starting small wasn’t simply that it reduced risk. It created evidence.

The company could observe what people actually did, what created value, where friction appeared and what was worth scaling. The first experiment did not need to prove the entire AI strategy. It needed to make the second decision better.

Three Decisions to Get Unstuck

A CEO who identifies with any of the above might consider three simple moves.

Decide who owns the next decision

Not permanently. Not perfectly.

Just choose someone accountable for moving the conversation forward and reporting back, and give them enough latitude to investigate properly.

You need not hand over broad decision rights, at least not yet. But someone needs responsibility for converting general interest into concrete questions, experiments and recommendations.

The goal is not to design the perfect AI governance structure on day one. It is to eliminate the ownership vacuum around what happens next.

Signal permission explicitly

If your people are hiding their AI use or non-use, they’re waiting for clarity from above.

A statement, a policy, a set of guardrails—anything that makes clear that thoughtful experimentation is legitimate and defines where the boundaries sit—can reduce friction and give leadership a more accurate picture of what is actually happening.

Permission is not the same as enthusiasm. It is the removal of ambiguity about what employees are allowed to try, disclose and learn from.

Pick one bounded starting point

Don’t begin by asking, “Where is the biggest AI opportunity in the company?”

Ask instead: Where can we learn something useful without making a large, irreversible bet?

A good first use case should be small enough to move, meaningful enough to teach you something and measurable enough to inform the next decision.

The objective of the first experiment is not transformation. It is better information.

Analysis paralysis breaks when the first move generates evidence rather than trying to settle the entire strategy in advance.

The Implication

If what I learned, and what ADKAR helped me see in those conversations, is in any way indicative of a broader trend, then the organizations making progress aren’t necessarily the ones with the most AI expertise.

At least not in the world I have access to.

They’re the ones that have figured out how to make a few basic organizational decisions: who owns the next move, what people are permitted to do, where learning should begin and how evidence from one experiment will shape the next.

The rest are waiting, or more accurately, stalled. Not because they don’t believe. Not because they aren’t ambitious. But because they haven’t resolved more basic questions: In what ways, if at all, could AI materially advance something important to us? Who owns the initiative? What does appropriate investment look like? What evidence would make the next decision easier?

“Even measuring productivity gains from AI is harder than it sounds,” said Michael Johnson, CEO of Metrc, a provider of cannabis tracking and regulatory systems. “Most of us don’t actually know how long a given task takes our workforce on average. Without that benchmark, how do you quantify the gain in real dollars?”

That’s a strategic question first.

And that may be the larger point.

Many organizations are behaving as though AI readiness is mainly a function of technical capability: better tools, more licenses, more training, more expertise.

But capability is only useful once an organization can decide what to do with it.

The companies that move may not begin with better answers. They begin by creating the conditions to produce better answers: visible behavior, clear permission, accountable ownership, bounded experiments and evidence that makes the next decision easier.

The AI gap, in other words, may increasingly be less about who has the best technology than who has built the better decision system around it.