McKinsey Says Spend Five Dollars on People for Every Dollar on AI Tools

If your company bought AI licenses this year and the productivity numbers didn't move, you're the rule, not the exception — and McKinsey's new data says the fix costs five times what you already spent on the tool.

The 6% and the 5:1 ratio

McKinsey's 2026 State of AI survey, published August 25 and drawing on 1,719 professionals worldwide, found that only 6% of organizations qualify as "AI high performers" — companies that can attribute 5% or more of EBIT to AI. That's despite 80% of individual employees reporting real productivity gains from using AI day to day. The gap between what individuals feel and what shows up on the balance sheet is the whole story: people are getting faster at tasks, but the organization around them isn't capturing it.

McKinsey's prescription is specific: for every dollar spent on AI technology, spend five dollars on the people who use it — training, workflow redesign, and the time it takes to figure out what actually works. Buying seats isn't the investment. Teaching people what to do with them is.

Two other data points landed the same week and point the same direction. Sam Altman told the David Senra podcast (Aug 23–24) that he was "too ambitious on timelines" for mainstream AI adoption, because "the economy just has so much inertia" — changing how people actually work is harder than it looks from inside an AI company. And Bill Gates published a 6,000-word essay (Aug 26) arguing that AI is closing an escape hatch prior waves of automation left open: the option to fall back on raw cognitive ability when a credential didn't matter yet. He cited Stanford payroll data showing a 16% relative decline in employment for 22-to-25-year-olds in AI-exposed roles.

What this means Monday morning

None of this is an argument for using AI less. It's an argument against the rollout McKinsey's 94% figure describes: give people a login, assume competence follows, and check back in a quarter. The skill that's missing isn't a prompting trick — it's the judgment to look at a task and decide whether AI actually fits it, before you spend budget assuming it does.

That's a trainable skill, and it's cheap relative to the tool spend McKinsey is describing. It looks like: before assigning a task to AI, ask what happens if the output is wrong, and who's accountable for catching it. Do that consistently and you're most of the way to the habit McKinsey's high performers already have.

Two teams, two outcomes

A marketing manager at a 40-person SaaS company got budget approval for AI writing tools in January. By August, the team was using them daily — for first drafts, subject lines, ad copy variations. But nobody had defined which decisions still needed a human sign-off, so drafts went out with claims nobody had checked and metaphors nobody would have chosen. The tool spend was real. The training spend was zero. Her account of the quarter was that the team got faster at producing more of the same problem.

Contrast that with an operations director at a 150-person logistics company, who spent the same tool budget but paired it with four hours of team time every month reviewing what worked and what didn't — which reports AI drafted well, which ones still needed a person to catch a wrong assumption about a client contract. Six months in, her team wasn't using AI for more tasks. They were using it for the right ones, and they could tell you which ones those were.

The takeaway

The 94% of companies not seeing AI returns didn't buy the wrong tool — they skipped the training that turns access into a result, and McKinsey just put a number on how much that training is worth.