AI Doesn't Have a Technology Problem. It Has an Execution Problem.

Adoption of AI in large companies is now close to universal. In McKinsey's 2026 State of AI survey of 1,719 respondents in 97 countries, nearly nine in ten said their organizations use AI regularly in at least one business function, and 80 percent said AI had improved their own productivity. Only 37 percent, however, reported that AI had contributed anything to their organization's earnings, roughly the same share as a year earlier. About 6 percent, the group McKinsey describes as high performers, attribute 5 percent or more of their EBIT to AI.

Access to technology does not explain this gap. Capable models are available to most companies, and the cost of using them continues to fall. The gap lies in what happens after a model is deployed: whether the work around it is redesigned, whether someone is accountable for the result, and whether the gains are converted into lower costs, higher revenue or capacity that is put to productive use. The same survey found that nearly three-quarters of high performers had fundamentally redesigned workflows because of AI, compared with about a quarter of other respondents.

Where programs lose track of value

The path from an AI model to a financial result has more stages than most programs measure. A model is tested in a pilot, deployed into production and adopted by users. To produce value, the workflow it supports then has to change, people have to work in the new way consistently, an operating measure such as cycle time or cost per case has to improve, and that improvement has to appear in a profit and loss line that someone owns.

Most program reporting stops partway through that sequence. Completed pilots, deployed systems and trained users are real milestones, and they are easy to count. They are also the stages at which a program can appear successful while the underlying work is unchanged. In several programs I have reviewed, the status reports were accurate and on track, yet front-line teams were still keeping manual workarounds alongside the new system, exceptions were still handled through the old process, and customer response times had barely moved.

An example from a large financial institution

In a previous role, I took over a digital and AI program at a large North American financial services group. The program was six months into a three-year plan worth roughly a billion dollars, spread across about five hundred initiatives. These covered data and platform work, automation in underwriting, claims and the contact center, and the consolidation of more than five thousand websites, apps and domains. All of the initiatives were reported as on plan, and the business case projected at least a dollar of savings for every dollar invested, with additional revenue on top.

Before proposing any changes, I asked each initiative lead to explain how a dollar of projected savings would actually be realized: which manual step would be removed, which queue would shrink, whose capacity would be released, and in which business unit's results the saving would appear. Most answers were reasonable estimates, but few had been tested against how the work was actually done. The benefits had been modeled at the portfolio level, largely without input from the operations leaders who would have to deliver them.

The program did not need a new strategy. It needed operating capabilities that the original plan had assumed were already in place. The most useful step was to agree on a single customer measure. Customers in every market reported that logging in and completing a transaction took more than ten seconds, against an internal standard of two. That measure gave the program a clear basis for prioritizing work and for declining work that did not contribute to it. Over three years, about four thousand websites that did not serve that goal were retired, more than five thousand digital properties were consolidated into ten main domains, login success rose above 99 percent, and customer satisfaction improved by double-digit points in the segments affected.

Three other changes were important. Digital and AI work became part of each business line's own plan rather than a separate program, which made business leaders accountable for results. The finance function agreed on what would count as a benefit before work was funded, and then tracked each benefit from early adoption measures through to the financial statements. Senior leaders also continued to support the work during periods when operating measures had improved but financial results had not yet followed. Over the three years, the program exceeded its financial and customer targets, although the plan changed several times along the way.

What leaders can do

Four practices help programs move from reporting activity to changing results.

1. Define progress in operating terms. For each initiative, ask which manual steps, reports, handoffs or queues have been removed or reduced. If nothing has been removed, AI is probably being added to existing work rather than changing it.

2. Give each outcome a business owner. The accountable person should run the part of the business where the results will appear and have the authority to change the workflow. Technology teams can deliver capability, but they cannot change how a business line operates.

3. Agree on benefit definitions with finance before funding. When finance helps set the baseline, the measures and the point at which a benefit is counted, later disputes about results become much less common, and the board is more likely to trust the numbers.

4. Track leading and lagging measures together. Adoption, exception rates and cycle times usually move months before financial results. Reporting both shows whether a program is on course before the financial effect is visible.

Conclusion

The gap between AI adoption and AI results is mainly an organizational issue. The companies that close it will not necessarily have better models than their competitors. They will have clearer ownership, redesigned workflows and a disciplined way of turning operational improvements into financial results. A useful starting point is to ask, for the largest AI initiatives in the portfolio, which line of the income statement has changed as a result and who is responsible for it.