Why Do CEOs See AI Value but Can't Scale It?

Nearly nine in ten CEOs now say their companies see some cost or revenue benefit from AI, almost entirely in targeted areas rather than at scale. That is the finding at the center of BCG's latest AI Transformation CEO Survey (n=152, July 2026), and it describes a specific kind of stuck: the pilots work and the value is real, yet it will not spread across the business. Why it stalls is the part most operators get wrong. The block lives in the organization those things are bolted onto, not in the model, the data pipeline, or the tooling.

Organizational barriers outrank the technical ones

When BCG asked CEOs what limits turning AI into financial impact at scale, the top two answers sat outside technology. The single most-cited barrier, at 56%, was an unclear link between AI initiatives and specific financial outcomes. The second, at 55%, was people, workflows, and incentives that had not been redesigned for AI. Technology and data capability gaps came third, at 49%: real, and still ranked below two problems that are organizational rather than architectural.

That ordering is the whole story. Most AI post-mortems reach for a technical explanation because a technical explanation is fixable with a purchase order. The data says the binding problems are the ones a purchase order cannot touch: nobody defined what number the initiative was supposed to move, and nobody changed how the work gets done around it. You can buy a better model. You cannot buy a redesigned workflow or a P&L owner accountable for the result.

The say-do gap: why self-report fails as diligence

One number here should stop an operating partner cold. In the same survey, 56% of CEOs cited the missing link between AI and the P&L as a key barrier, yet only 14% said they define a P&L impact for every AI initiative before it launches. That is a 42-point gap between the CEOs who name the problem and the ones who have done anything about it.

Read that gap carefully, because it is the reason a management team's self-assessment is worthless as diligence. A CEO who tells you AI value creation is a priority is, on this data, four times more likely to be describing an aspiration than a practice. The recognition is nearly universal; the discipline is rare. The same pattern repeats across the survey: 64% of companies run AI pilots, but only 26% embed AI into a broader business transformation. And only 30% include HR in AI governance, against 82% who include technology, which tells you the workforce redesign everyone says matters is, in practice, nobody's job.

When self-report and behavior diverge by 42 points, treat the self-report as a claim awaiting proof. The only thing that counts is what is built: defined value paths, accountability with a real owner, roles redesigned around the work. All of it observable rather than surveyable.

Where the value lives: 10 / 20 / 70

BCG's own experience with clients puts a proportion on it: roughly 10% of the value from AI comes from the algorithms, 20% from the data, and 70% from changes to the operating model and new ways of working. That figure is BCG's field observation rather than a survey result, and it lines up exactly with what the CEO data implies. If 70% of the value is organizational, then a program that spends 90% of its energy on models and data is optimizing the 30% and starving the 70%.

The high performers in the survey behave accordingly. CEOs whose companies capture significant AI value are far more likely to have put the foundation in place first: 46% restructure accountability for results versus 24% of laggards, 25% track financial value straight to the P&L versus 5%, and 44% fully fund people and change management versus 24%. And they are roughly seven times more likely to redesign workflows end-to-end. The gap between the companies getting value and the ones stuck in pilots is almost entirely a gap in organizational readiness.

What this means inside a hold period

BCG surveyed generalist large-enterprise CEOs rather than PE-backed portfolio companies. But the diagnosis travels, and inside a finite hold period it gets sharper. A portco does not have five years to discover that its AI spend is landing on the 30% that does not compound. If the operating model, the data, the process, and the talent underneath AI are not ready, then funding AI acceleration first buys expensive pilots that stall exactly where the survey says they stall, rather than the value creation the budget promised.

This is where our diagnosis and our prescription part company from the data. BCG's numbers validate the diagnosis: organizational debt, more than technology, is what binds AI value. The prescription is ours: resolve the binding constraint before you fund acceleration. Every enterprise carries some mix of four pre-existing debts (data, technology, process, and talent), and one of them is the constraint that caps the return on everything else. Pouring AI budget over an unresolved talent or process debt is the mechanism behind the 42-point say-do gap. The work is to find the binding constraint, fix it, and only then let AI compound on top of a foundation that can carry it. (For a board's role here, separate BCG research found nearly two-thirds of chief transformation officers said their boards were mostly limited to status updates; passive oversight is its own kind of unresolved debt.)

That sequencing, constraint first and acceleration second, is the difference between the 26% who embed AI and the 64% who keep running pilots.

Start by naming the constraint

You cannot resolve a binding constraint you have not identified. The Enterprise Debt Index scores the four pre-existing debts (data, technology, process, and talent) and tells you which one is the constraint on AI in your business, and what your deal type calls for first. It is twelve questions, and it is the read that should come before any AI acceleration budget gets approved. From there, the AI Value Creation framework sequences the initiatives that a ready foundation can carry.