The two numbers from McKinsey’s 2025 study sit awkwardly next to each other. The first is 88%, which is the percentage of organizations currently using AI in at least one part of their business. The second is 39 percent, which could indicate a measurable impact on revenue. The distance between these numbers is the real story of where enterprise AI will stand in the second half of 2025.
This figure is based on a McKinsey AI report published on November 5, 2025, which drew 1,993 respondents from 105 countries surveyed in Northern Summer. “88% report regularly using AI in at least one business function, compared to 78% a year ago,” the report’s authors wrote. These are self-reported numbers by survey respondents, not an audit of every company on the planet. Also, keep in mind that these numbers represent usage elsewhere in your organization, not usage that changes how your organization works.
Adoption is not the same as influence
Near-universal adoption is easily misinterpreted as near-universal change. Research suggests otherwise. The authors of the report are frank: “But at the enterprise level, the vast majority are still in the experimental or pilot stage.” Many teams are running models in their corner of the business. Far fewer companies have rewired their businesses around it.
The scaling number makes the gap more specific. McKinsey estimates that only about one-third of organizations have begun scaling AI across the enterprise, and nearly two-thirds are still doing so.
Only 7% report that their AI is fully scaled. When it comes to benefits, around 39% of respondents believe that the impact on company-level EBIT is due to AI, with most pegging that number at less than 5%. The authors are not embellishing it. “While it remains rare for the use of AI to have a meaningful impact on company-wide revenue, our findings suggest that thinking big can pay off.”
pilot’s trap
This pattern is not unique to one study. An independent MIT Project NANDA report in August 2025, led by Aditya Challapally, reached conclusions along similar lines in a different way, finding that “the 95% failure rate of enterprise AI solutions is the clearest demonstration of the GenAI divide.” This headline figure is disputed and is based on one study, which narrowly defines “failure” as no rapid revenue or P&L impact. By the report’s own definition, we read that approximately 95% of pilots had no measurable revenue impact, which is consistent with McKinsey’s more conservative version of the same findings.
Why do so many initiatives stall at the testing stage?
A pilot can be executed with the enthusiasm of one team and a shoestring budget. Scaling requires redesigned processes, retrained staff, leadership willing to own the results, and a tolerance for disruption that is never tested in a contained experiment. A model that impresses in demos must survive contact with messy data, existing systems, and people changing its work. Most of the friction is likely to be there, not in the model.
that the minority does things differently
Approximately 6 percent of respondents met McKinsey’s “AI High Performer” criteria. This means that AI is contributing significant value to more than 5 percent of EBIT, according to the report. What separates the two is less about smarter algorithms and more about how deeply they’ve rebuilt themselves around their tools. The authors state that “half of AI talent intends to use AI to transform their business, and most are redesigning their workflows.”
Workflow redesign is what keeps reports coming back. High performers are not limited to automating existing processes. They are “rethinking from the ground up” and incorporating AI into their workflows and decision-making rather than building it into what they already have. 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 study found across the variables it tested, but a single year of study data does not allow us to distinguish between cause and correlation.
McKinsey reports that nearly 80% of respondents set efficiency as a goal for their AI work, while high-performing companies also pursue growth and innovation. Doing the same thing a little cheaper is a smaller goal than doing something completely different, and research associates bigger goals with bigger rewards.
Questions left by the investigation
What the high performer survey results cannot settle is the direction of the arrow. Companies looking to redesign and transform their workflows may be able to move forward with these choices. Or perhaps companies that are already moving forward simply have the resources and confidence to take on bigger challenges. The report does not claim to solve this. Top-performing companies are also devoting a much larger portion of their digital budgets to AI. This is both a recipe that others can follow and an indicator for organizations that already have room to spend.
According to these numbers, the gap between 88% adoption rate and 39% measurable impact does not indicate that AI has failed. This shows that most organizations are still in the early stages, buying access to tools is the cheap part, and re-engineering how work actually gets done is being done by a few and most organizations still not doing it. Next year’s research will show whether the gap between adoption and impact is closing.
