McKinsey senior partner David Pralong says the numbers do not justify the hype.
McKinsey’s latest research finds that 89 percent of organizations used AI in at least one business function this year, up from 88 percent a year ago. Yet the share of high performers, defined as companies attributing at least 5 percent of earnings before interest and taxes to AI and reporting value from its use, stayed flat at 6 percent. Thirty-seven percent report some positive effect on earnings, but the gap between broad use and real payoff has not closed.
Pralong argues that many executives are confusing narrow successes with a scalable strategy. He points to a study of about 5,200 customer-support agents, where AI assistance increased issues resolved per hour by 15 percent, and randomized trials involving about 4,900 software developers, where AI coding assistants helped complete about 26 percent more tasks. Those are meaningful gains, he writes. They do not prove that adding AI to any business process will improve earnings.
The reason, he says, is that contact centers and software teams already had structured workflows before the models arrived. Contact centers had queues, tracked outcomes, and stored prior interactions. Software teams already used testing, continuous integration, and code review. AI entered systems that were already built to measure and correct work.
Pralong says the harder problem is redesigning work that was never organized around a clear outcome or a way to verify success. He uses a bank as an example: AI might read documents faster for small-business loans, but if an application still moves through handoffs among sales, credit, compliance, and operations, the customer may see little improvement. A quicker first step does not remove the delays and errors later in the chain.
His prescription is managerial, not magical. The company has to decide what decision it is trying to make, what evidence is required, who has authority, and which exceptions need specialist review. AI can help gather and check evidence, but a named team has to own the process from submission to decision. He says the right measures are time to decision, error rates, and the loans the bank can responsibly serve, not just hours saved.
Pralong also says leadership must answer questions about authority, risk, and what employees will do with time regained. Without that, a promising pilot can remain a pilot. Employees also notice the uncertainty and ask whether they are training their replacement.
McKinsey’s own research published in July adds to that caution. Among leaders in the earliest stage of AI adoption, organizations whose workflows had been redesigned were 5.3 times as likely to report enterprise-level value as those whose workflows had not: 32 percent versus 6 percent.
The market has already voted on the easy story. Investors and executives alike want to believe software alone can fix productivity. But the numbers point to a less convenient truth: capital rewards clear rules, not demos. AI can accelerate existing processes, but it does not replace leadership, accountability, or the discipline required to rebuild work around measurable results.
