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 central finding of BCG's latest AI Transformation CEO Survey (n=152, July 2026). The pilots work and the value is real, but it does not spread across the business. Most operators look for the cause in the model, the data pipeline, or the tooling. The survey places the cause in the organization around those tools.

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%. Those gaps are real, and CEOs still ranked them below two organizational problems.

Most AI post-mortems reach for a technical explanation, because a technical problem can be fixed with a purchase order. The survey points to problems a purchase order cannot fix. Nobody defined what financial result the initiative was supposed to produce, and nobody changed how the work gets done around it. A company can buy a better model, but a redesigned workflow and a P&L owner accountable for the result have to be built inside the business.

Why management self-assessment is weak diligence evidence

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. On this one measure, 42 points separate the CEOs who name the problem from the ones who have done anything about it.

This gap is why a management team's self-assessment cannot be used as diligence evidence. On this data, a CEO who tells you AI value creation is a priority is four times more likely to be describing an aspiration than a practice. The same pattern repeats across the survey. 64% of companies run AI pilots, but only 26% embed AI into a broader business transformation. Only 30% include HR in AI governance, against 82% who include technology. In most surveyed companies, no function owns the workforce redesign that CEOs say matters.

When self-report and behavior differ by 42 points, treat the self-report as a claim that still needs evidence. The evidence is what the company has built: defined value paths, accountability with a named owner, and roles redesigned around the work. Each of these can be observed directly in the business rather than asked about in a survey.

Where the value comes from: 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 is consistent with what the CEO data implies. If 70% of the value is organizational, a program that spends 90% of its effort on models and data is working on the 30% and underfunding 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 directly to the P&L, versus 5%. 44% fully fund people and change management, versus 24%. They are also roughly seven times more likely to redesign workflows end-to-end. The difference between the companies getting value and the ones stuck in pilots is almost entirely a difference in organizational readiness.

What this means inside a hold period

BCG surveyed generalist large-enterprise CEOs rather than PE-backed portfolio companies. The diagnosis still applies to portcos, and a finite hold period makes it more pressing. A portco does not have five years to discover that its AI spend is going to the 30% of value that does not compound. In the AI Cost Optimization Framework's six-layer model of AI spend, the three least visible layers (people, waste, and risk) account for 55% of the total. Governance is the layer built to detect that spend. If the operating model, the data, the process, and the talent underneath AI are not ready, funding AI acceleration first produces expensive pilots. Those pilots stall where the survey says they stall, and the value creation the budget promised does not arrive.

BCG's numbers support the diagnosis: organizational debt, more than technology, is what limits AI value. The prescription goes beyond the data, and it is ours: resolve the binding constraint before you fund acceleration. Every enterprise has 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. Spending AI budget on top of an unresolved talent or process debt is what produces the 42-point gap between stated priorities and practice. The work is to find the binding constraint, fix it, and only then add AI to a foundation that can support it. (On the board's role, BCG's The Secret to a Successful Transformation? An Engaged Board found nearly two-thirds of chief transformation officers said their boards were mostly limited to status updates. We treat passive board oversight as a fifth debt.)

In our view, companies that resolve the binding constraint first are more likely to move from pilots to embedded AI. The survey does not test this directly.

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). It 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 should come before any AI acceleration budget gets approved. From there, the AI Value Creation framework sequences the initiatives that a ready foundation can support.