Most Companies Still Can't Point to AI Profits, McKinsey Finds

Key points

• Only 6 percent of the 1,719 professionals McKinsey surveyed in 2026 qualify as AI "high performers," the same share as in 2025.
• 37 percent of respondents say AI has contributed at least something to their EBIT (earnings before interest and taxes, a standard measure of profit), flat year-on-year.
• 80 percent of individual AI users say the technology improved their personal productivity, but those gains have not yet shown up in company-wide finances.
• 39 percent of respondents expect their employer to cut jobs because of AI in the coming year, up from 32 percent in 2025.
• 20 percent of respondents say AI running costs have actively limited how much of it they can use.

McKinsey's annual State of AI report, covering 2026, carries an optimistic headline: businesses are "on the road to ROI." Read past the title, though, and the numbers tell a more cautious story.

ROI stands for return on investment, the basic question of whether you get more money back than you put in. For most companies surveyed, the answer is still: not clearly, not yet.

So who is actually making money from AI?

Very few. McKinsey defines an AI "high performer" as an organisation where AI accounts for at least 5 percent of EBIT and leaders describe its impact as significant. Just 6 percent of respondents met both conditions, the same figure as last year.

A broader 37 percent say AI has contributed something to profits. McKinsey itself concedes that share is "about the same" as 2025. Flat, in other words, despite another year of heavy spending.

What is everyone actually using AI for?

Agents are the new focus. An AI agent is software that can carry out multi-step tasks on its own, going further than a simple chatbot. Among companies with more than $1 billion in annual revenue, 40 percent say they are now scaling AI agents across their business, up from 27 percent in 2025.

Coding agents, tools that write or assist with software code, are also climbing. Nearly a third of respondents say their company chose to build software in-house using these tools rather than buy an existing product.

That shift carries a practical risk McKinsey does not dwell on: AI-generated code can contain bugs or security gaps that take skilled developers time to find and fix.

Should employees worry about their jobs?

Worry is spreading, though the reality has repeatedly lagged behind the fear. In 2025, job cuts from AI "fell well short" of what the previous year's respondents had predicted, McKinsey notes. History, then, counsels some scepticism.

Still, the direction of expectation is clear. This year, 39 percent of respondents expect their employer to cut headcount because of AI in the next twelve months. Last year that figure was 32 percent. A separate 43 percent still expect little or no change.

For workers, the honest read is: watch what your employer actually does, not what survey respondents think others will do.

What does this mean for ordinary people at work?

Eighty percent of employees who use AI in their day-to-day roles say it improved their personal productivity. That is the clearest win in the entire report.

The problem, as The Register first reported, is that individual productivity gains and company-wide profit growth are two very different things. A nurse who documents faster or a shop owner who drafts emails quicker may feel the benefit. Their employer's balance sheet may not, at least not yet.

Costs are also a real limit. One in five respondents said AI running costs, the fees for using these services, have actually constrained how widely their organisation can deploy the technology.

Common questions

Does this mean AI is failing?

Not exactly. Individual users report real productivity gains, and adoption is rising. The gap is between what companies expect AI to do for profits and what it has demonstrably done so far.

When might companies see clearer returns?

McKinsey's report does not give a timeline. More respondents than last year believe AI will reshape their business within three years, but that is expectation, not evidence.