88% of Companies Use AI. 6% Actually Profit From It.

88% of organizations now use AI somewhere in the business. McKinsey's own State of AI research puts that number in writing. It should be the least controversial fact in enterprise technology. Nearly every company checked the box. Now look at the number sitting right next to it. Only about 6% qualify as genuine AI high performers. That means they can point to real, attributable earnings impact of more than 5%. Adoption is nearly universal. Profit from it is rare.

I hear versions of this same mix-up in almost every executive talk I have right now. A CFO tells me the company "uses AI everywhere." Then I ask a simpler question. Which line on the income statement moved because of it? Most of the time, the room goes quiet. Or someone points to a productivity guess that was never checked against a real dollar figure. The CFO isn't at fault here. A tool got adopted faster than anyone built the measurement for it. That gap shows up in exactly this moment.

McKinsey's data shows exactly where the gap opens up. That 88% figure is up from 78% just a year earlier. Adoption moved fast. 62% of companies are experimenting with AI agents specifically. But nearly two-thirds of all companies surveyed haven't started scaling AI across the enterprise at all. They're still running pilots, department by department, with no plan to connect them into something the whole company depends on. Only 39% can attribute any EBIT impact to AI at all, and that's the generous number. The 6% of real high performers is the number that actually separates winners from everyone else.

A bigger budget doesn't split that 6% from the rest. A smarter model doesn't either. McKinsey found high performers are nearly three times as likely as everyone else to fundamentally redesign the workflow itself, not just add a tool on top of the old one. Researchers called it one of the strongest single factors behind real business impact. The tool was never the differentiator. Rebuilding the process around it was.

I ran a $1.1B innovation portfolio across 14 countries at Cisco. I watched this exact pattern play out with technology after technology, long before generative AI existed. A new capability shows up. Most teams bolt it onto the existing workflow, because that feels faster and lower risk. A smaller number of teams stop. They redraw the actual process and accept the short-term disruption that comes with doing it right. That second group takes longer to launch. Then it moves faster than everyone else afterward. McKinsey just measured that same choice at scale, across thousands of companies, with real earnings data behind it.

There's a risk hiding underneath the scaling gap that deserves its own attention. 51% of companies in this same research had at least one real bad outcome from their AI use. Here's the detail that matters most. High performers are far more likely than everyone else to have a defined process for deciding when an AI output needs a human to check it before it ships. McKinsey named this one of the top factors separating the 6% from the rest. Speed and risk don't separate the companies capturing real value from everyone else. A checkpoint built before scaling does, not one added after something breaks in production.

I think about the railroads laying track across a continent in the 1800s. Track alone didn't create the value. Standardized schedules did. Safety signals did. A system for coordinating hundreds of trains on the same lines did. The rails came first and looked like progress on their own. The real economic gain showed up only once someone built the operating system running on top of the rails. Enterprise AI is at the rails stage for most companies right now. The 6% already started building the operating system.

The relevance cliff sits directly underneath this scaling gap too. Every team stuck running a pilot instead of a redesigned workflow keeps using old skills on a new tool. Those skills expire faster the longer that mismatch continues. A company can widen the gap between its 6% and its 94% without meaning to. Letting pilots run on and on instead of forcing the harder redesign work is how that happens.

Before your company announces another AI pilot, ask the one question that actually predicts which group you'll land in. Are you willing to redesign this workflow from the ground up, with a real review step built in? Or are you hoping the tool does the redesign work for you? McKinsey's data already knows which answer leads to the 6%.

Sources: McKinsey & Company, "The State of AI in 2025: Agents, Innovation, and Transformation," November 5, 2025.