AI is everywhere. Value is not: what the latest evidence says

By almost any measure, AI adoption is no longer the story. The story is what happens after adoption. Across industries, organisations have deployed assistants, copilots and pilots at speed, and boards are now asking a more demanding question: where is the return?

The most cited survey evidence answers it candidly. Adoption is broad. Value is narrow. And the gap between the two is largely about management, not machines.

88%

of organisations use AI in at least one function (McKinsey, 2025)

39%

report any effect on enterprise-wide EBIT, mostly under 5%

~6%

are "AI high performers" attributing over 5% of EBIT to AI

Wide adoption, thin returns

McKinsey's State of AI survey, reported in November 2025, found that 88% of organisations use AI in at least one business function, up from 78% the year before. Yet nearly two-thirds were still experimenting or piloting rather than scaling across the enterprise. Sixty-two per cent were at least experimenting with AI agents, and 23% were scaling them in one or two functions.

On financial impact, 39% reported any effect on enterprise-wide EBIT, and most of those attributed less than 5% of EBIT to AI. Larger shares reported softer gains: 64% said AI improved innovation, and almost half cited improvements in customer satisfaction and competitive differentiation.

What the top few percent do differently

The small group McKinsey labels high performers shares recognisable traits. Their senior executives are three times more likely to own AI initiatives personally. More than a third dedicate over 20% of their digital budgets to AI, against under 10% for other organisations. They are three times more likely to expect transformative change within three years. And they are more likely to redesign workflows around AI rather than lay AI on top of existing ones.

None of these is a technology choice. They are decisions about ownership, budget and process design, which is why the same tools produce very different outcomes in different organisations.

The hype layer: agents and agent-washing

Gartner's June 2025 prediction is a useful corrective. It expects over 40% of agentic AI projects to be cancelled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls. Gartner also warns of "agent washing", the rebranding of assistants, robotic process automation and chatbots as agentic, and estimates only about 130 of the thousands of vendors claiming agentic capability are real.

“Most agentic AI projects right now are early stage experiments or proof of concepts, often misapplied.”

The same Gartner release is not pessimistic about the direction. It expects at least 15% of day-to-day work decisions to be made autonomously by agentic AI by 2028, up from 0% in 2024, and a third of enterprise software applications to include agentic AI by 2028, from under 1% in 2024. The prediction is not that agents fail; it is that unscoped, unmeasured agents will.

A disciplined way to begin

The organisations we advise are rarely short of ideas. They are short of a method for choosing and finishing. We suggest a five-step discipline.

• Choose one process with a clear owner, a measurable cost and enough volume to matter, such as invoice handling, first-line support or a recurring report.
• Map how the work really flows today, including the exceptions that people handle informally.
• Redesign the process around the capability, deciding which steps AI performs, which a person reviews and which are eliminated.
• Set the measure before the build: cost per transaction, turnaround time or error rate, with a baseline.
• Run it for a quarter, then decide: scale, adjust or stop. Stopping a project that does not pay is a result, not a failure.

What this means for growing organisations

For mid-market and growing firms, the evidence is good news. The advantage of the leaders is not scale or budget; it is clarity about where AI fits and the willingness to redesign the work. A smaller organisation can decide faster, own the initiative at the top and measure one process well. The practical constraint is expertise, which Celonis's 2026 research puts at the leading hurdle (47% of leaders), and which can be addressed by building capability deliberately rather than buying it indefinitely.