Two numbers from McKinsey’s 2025 survey sit awkwardly next to each other. The first is 88 percent, the share of organisations now using AI in at least one part of the business. The second is 39 percent, the share that can point to any measurable effect on the bottom line. The distance between those figures is the real story of where corporate AI stands in late 2025.
The numbers come from McKinsey’s State of AI report, published on 5 November 2025, drawing on 1,993 respondents across 105 countries surveyed over the northern summer. The report’s authors write that “88 percent report regular AI use in at least one business function, compared with 78 percent a year ago.” Keep in mind that these are self-reported figures from survey respondents, not an audited count of every company on earth, and they describe usage somewhere in the organisation, not usage that has changed how the organisation works.
Adoption is not the same as impact
Near-universal adoption is easy to misread as near-universal transformation. The survey suggests otherwise. The report’s authors are blunt: “But at the enterprise level, the majority are still in the experimenting or piloting stages.” Plenty of teams are running a model in one corner of the business; far fewer have rewired the business around it.
The scaling figures make the gap concrete. By McKinsey’s reckoning, only about one-third of organisations have begun to scale AI across the enterprise, leaving nearly two-thirds that have not.
The share reporting that AI is fully scaled is just 7 percent. On profit, around 39 percent of respondents attribute any enterprise-level EBIT impact to AI, and most of those put the figure below 5 percent. The authors do not dress it up: “Meaningful enterprise-wide bottom-line impact from the use of AI continues to be rare, though our survey results suggest that thinking big can pay off.”
The pilot trap
The pattern is not unique to one survey. An independent MIT Project NANDA report from August 2025, led by Aditya Challapally, reached a directionally similar conclusion by different means, finding that “The 95% failure rate for enterprise AI solutions represents the clearest manifestation of the GenAI Divide.” That headline figure is contested and rests on one study, with “failure” defined narrowly as no rapid revenue or profit-and-loss impact. Read as roughly 95 percent of pilots showing no measurable bottom-line effect by the report’s own definition, it lines up with McKinsey’s quieter version of the same finding.
Why do so many efforts stall at the pilot stage?
A pilot can run on a single team’s enthusiasm and a modest budget. Scaling demands redesigned processes, retrained staff, leadership willing to own the result, and a tolerance for disruption that a contained experiment never tests. The model that impresses in a demo still has to survive contact with messy data, existing systems, and the people whose work it is meant to change. Most of the friction likely lives there, not in the model.
What the small minority do differently
Roughly 6 percent of respondents clear McKinsey’s bar for “AI high performers,” meaning AI drives 5 percent or more of EBIT plus what the report calls significant value. What separates them is less about smarter algorithms than about how deeply they rebuild around the tools. The authors note that “Half of those AI high performers intend to use AI to transform their businesses, and most are redesigning workflows.”
Workflow redesign is the factor the report keeps returning to. High performers do not limit themselves to automating existing processes; they “rethink them from scratch”, building AI into workflows and decision-making rather than bolting it onto what already exists. The same group is also three times more likely to use AI to drive transformative change rather than narrow efficiency gains. This is perhaps the strongest signal the survey found across the variables it tested, though a single year of survey data cannot separate cause from correlation.
McKinsey reports that around 80 percent of respondents set efficiency as an objective for their AI work, while high performers are also chasing growth and innovation. Doing the same things slightly cheaper is a smaller ambition than doing different things entirely, and the survey associates the larger ambition with the larger payoff.
The question the survey leaves open
What the high-performer findings cannot settle is the direction of the arrow. Companies that redesign workflows and aim for transformation may pull ahead because of those choices, or companies already pulling ahead may simply have the resources and confidence to attempt bigger things. The report does not claim to resolve this. High performers also commit a far larger slice of their digital budgets to AI, which is as easily a marker of an organisation that already has room to spend as it is a recipe others could follow.
The gap between 88 percent adoption and 39 percent measurable impact is not, on this reading, a sign that AI has failed. It is a sign that most organisations are still early, that buying access to a tool is the cheap part, and that rebuilding how work is actually done is where the small minority has gone and most have not yet. Next year’s survey will show whether the gap between adoption and impact is narrowing.
