"AI initiatives fail not because of technology but because leaders apply weak transformation discipline."

So begins BCG's recent article "Five Barriers CEOs Must Overcome for AI Impact." However, instead of then making a fairly familiar argument about change management, the author, Matthieu Berthion, says that the variable separating the companies generating real AI value from the ones who are stuck is the discipline applied by the chief executive.

This is pretty fascinating when you think about it and feels directionally correct to me. What many CEOs are doing right now is delegating their AI transformation work to other people. They get excited about the possibilities promised by AI, and then immediately hire a chief AI officer. And this, ironically, is exactly where the problem begins to arise.

The CEO then stops worrying about AI transformation and is no longer personally accountable for it. Instead, the chief AI transformation officer becomes the person spearheading the change—and while that's certainly better than doing nothing, it's not nearly as effective as it could be.

Sometimes, it simply exacerbates the problem.

What the Research Shows

Berthion calls out five barriers to success in the BCG piece, and the through-line across all of them is the same. Workflows are not getting redesigned. Incentives are not getting rewritten. The cross-functional trade-offs are not getting resolved. Even though pilots are running and things may look okay from the outside, the structural work that actually produces P&L impact is not happening. This is because real change requires the CEO to push the transformation from the top.

Taking this even further, Microsoft's 2026 Work Trend Index, published earlier this month, arrives at the same finding from a different angle. Twenty thousand workers, ten markets, trillions of productivity signals, and the headline result is that the organizational environment around a worker—meaning culture, manager support, and talent practices—accounts for sixty-seven percent of AI's reported impact. Individual mindset and behavior accounts for only about thirty-two percent. This suggests that the system around the person matters more than the person herself.

What both reports are effectively saying, in their own way, is that the rate-limiting input for AI value is leadership. It's not skills or tools or budget. It's leadership, which means the act of delegating AI is ultimately where everything breaks down.

The Avoidance

The reason CEOs do this is not hard to figure out. AI transformation, done seriously, requires the chief executive to do things the role has not historically required. It requires fluency they don't have. It requires personal modeling in front of their team. It requires rewriting incentives they wrote themselves. It also requires authorizing structural changes that can require staffing cuts. These are uncomfortable demands, and the easiest thing to do is to hire someone else and make them responsible. The chief AI officer is, at least in part, a way for the CEO to gesture at the AI transformation work they want to do without actually doing it.

But the work is the work, and there is no version of this transformation that succeeds without the CEO doing the hard things.

What Can't Be Delegated

A few responsibilities can’t be delegated.

The first is using AI personally and visibly. People watch what the CEO actually does, not just what they say in the all-hands meeting. A CEO who isn't using AI in their own work, in front of their team, is telling the company, without saying it, that AI is something other people deal with.

“We are an AI-first company" is meaningless if it comes from a CEO who never opens Claude, Gemini, or ChatGPT.

The second is changing what the company rewards. Microsoft found that only thirteen percent of AI users say they get rewarded for reinventing how they work with AI. Think about what that means. People are being told to change how they work, and almost none of them are being recognized for doing it. So most people keep doing the things they used to do, because that's what the bonus, the review, and the promotion still measure. Nobody but the CEO can change that. Until the things the company celebrates actually change, the behavior won't either.

The third is forcing the cross-functional changes that nobody else can. BCG estimates that seventy percent of AI's value sits in redesigning how work flows across teams, not in making any single team faster. Those redesigns are hard precisely because the cost lands in one team and the benefit lands in another. Department heads cannot fix this on their own. They do not have the authority or the political cover. The CEO does. The CEO is the only person who can stand up and say, we are redesigning this end-to-end, and here is who absorbs the cost. Without that, the seventy percent never gets unlocked.

The fourth and final thing that can’t be delegated is funding the work past the early wins. The first six months look great. Pilots ship. Productivity ticks up. Dashboards show real numbers. It feels like the transformation is working. However, the real work, the deeper integration that actually moves the P&L, sits past that point. BCG describes what happens next as a broken cash flow curve: companies cut investment right when they should be doubling down, because the early visibility makes the transformation look done. The CEO is the only person who can hold the line on funding the part of the work that does not yet show up on the dashboard.

A Useful Test

A CEO actually leading an AI transformation should be able to describe, in concrete terms, what they have personally done this quarter on each one. Not what their chief AI officer has done. Not what the transformation office has reported. What they did. If the only answer is sponsorship, the role has not yet moved.

AI did not arrive as a new initiative on top of the old job. It arrived as a new job. The strategy doc is not the strategy. The behavior of the chief executive is the strategy, and the rest of the company is watching.