In August, QuantumBlack, McKinsey's AI arm, published The State of AI in 2026: On the Road to ROI, based on a survey of 1,719 executives across 97 countries, fielded between May 4 and June 8, 2026. The paper deserves one caveat upfront: McKinsey sells AI advisory and implementation work to the same companies it surveys, so it has a commercial interest in the story of pending opportunity. Even so, its central numbers have been picked up without objection by TechTarget, a trade outlet with no such interest, and together they paint a picture that is hard to argue with. Adoption has leveled off, with 89% of companies using AI in some function versus 88% in 2025, while scaling across the enterprise keeps climbing, from 38% to 44%. What is not moving is the money: only 37% of organizations attribute a positive effect on profit to AI, the same share as a year ago, and only 6% fall into what McKinsey calls the high-performer category, defined as attributing at least 5% of EBIT to AI with a meaningful effect. That 6% has not moved in two years.
The gap sits not in the technology but between the individual and the company's books. 80% of respondents say AI makes them personally more productive and 50% say it helps them make better decisions, feelings that anyone who uses these tools daily will recognize. But that individual gain is not reaching the company's bottom line, except in that 6%. TechTarget sums it up in a useful image: adoption is sprinting ahead while financial return is barely crawling.
What sets that 6% apart
Per TechTarget's analysis of the report, the trait that most separates high-performing companies is having redesigned the workflow around AI, something close to three in four of them have done, versus roughly one in four of everyone else. The gain does not come from handing people a tool to use however they like on top of the same old process; it comes from changing the process itself. Size matters too, though unevenly: among companies with more than $1 billion in revenue, 40% are already scaling AI agents across the organization, up from 27% a year ago, while smaller companies are stuck flat at 22%, likely because they have less dedicated budget and staff to sustain a redesign. Cost, meanwhile, still holds back about one in five organizations, which suggests part of the bill for preparing data, integrating systems and keeping AI running is not being counted when the return gets calculated.
What to do before buying more seats
- Decide upfront which line on the books has to move. Before expanding a contract, write down which revenue, cost or time figure needs to change, and by how much. If nobody can name that figure, what is being bought is a feeling of productivity, not profit.
- Redesign the process, don't paste AI on top of the old one. Handing people a tool to draft the same thing faster rarely changes the outcome. The jump shows up when a whole step gets removed from the process, or the order of the work changes, not when one step that stays the same simply gets faster.
- Commit fixed budget and people, not a one-quarter experiment. The large companies in the report get further because they treat AI as a line of business with dedicated people, not a side trial. A small company can copy that logic at its own scale, naming an owner with real hours on the calendar, even if it cannot copy the budget.
- Count the full cost before promising a return. Include preparing the data, integrating the systems, training people and keeping the agent running, not just the license. That is the difference between the 37% that sees any effect and the 6% that sees a meaningful one.
On our own team, we do not count it as a gain when an agent works faster; we count whether a task that used to need a whole person now takes less human time once it is deployed, and that is the figure we show a client before expanding any agent.