91% of Companies Use AI Now. Only 6% Have Actually Made Money From It.

Adoption and impact are two very different numbers, and most of the AI-anxiety headlines quietly conflate them.

Per McKinsey's own tracking, 91% of businesses now use AI in at least one capacity, up from 78% two years ago, and 65% use generative AI specifically in at least one business function -- double the rate from ten months earlier. Those are the numbers that get quoted. The numbers that don't get quoted nearly as often, from the same research: only about a third of organizations have scaled AI beyond pilot programs, two-thirds are still in some form of experiment or test phase, only 39% report any EBIT impact they can attribute to AI at all, and just 6% qualify as what McKinsey calls AI high performers -- more than 5% of earnings attributable to it.

Read those together and the picture is very different from "AI has taken over the enterprise." It's closer to: almost everyone has kicked the tires, a third have put it to work at scale, and a small single-digit sliver have turned it into real, measurable money. The adoption number tells you the tools are in the building somewhere. The scale number tells you how much of the actual workflow has been rebuilt around it, which is what determines whether your specific role changes this year or three years from now. The performance number tells you how few companies have converted use into money as opposed to activity -- a lot of confident "we're going all-in on AI" announcements are coming from inside that 94% that hasn't cracked it yet.

The trap is reading "nothing's changed for my team yet" as "nothing's going to." The data supports a slower, lumpier rollout than the coverage suggests, not a reversed direction -- every quarter, more of that two-thirds moves from pilot to scaled, and the gap tends to close fast once a handful of competitors in an industry prove out the case, since that creates a board-level mandate for everyone else to catch up. The companies in the thin 6% high-performer slice concentrate in a few functions first -- customer service, software engineering support, marketing content production, data analysis and reporting -- which line up closely with the roles that show up as high or moderate exposure in task-level AI research.

McKinsey's State of AI survey data on enterprise adoption, scaling, and EBIT impact (2026), read alongside Job Resiliens' own AI Job Risk Index research.

Also worth reading: AI vulnerability score by job title · how we score AI resilience