The Real Numbers on Enterprise AI Adoption: Reading the Reports Side by Side

Two numbers side by side are this article's whole subject. In McKinsey's survey published in November 2025, 88% of companies regularly use AI in at least one business function — while only about a third have begun to scale it. The first number explains why everyone around you talks about AI; the second explains why so much talk has changed so little actual business.

The State of AI survey covered 1,993 respondents across 105 countries, one of the larger samples in enterprise AI research. Below, the key figures one group at a time — what each says on its own, and what they say together.

88% versus about a third: using is not scaling

88% of companies use AI regularly in at least one function, ten points up from 78% a year earlier; more than two thirds use it in multiple functions, and half in three or more. Yet the same survey finds only about a third beginning to scale. The distance between those figures is the distance between trialling and operating: tools bought, accounts opened, some staff using them — versus AI written into standard process, with permissions, acceptance criteria and an owner.

39% and under 5%: the P&L impact is still thin

The more sobering set concerns money: 39% of respondents report enterprise-level EBIT impact from AI — and note this is already the has-impact bucket — with most of them putting the effect below 5%. In other words, six in ten companies cannot yet name AI's contribution to profit, and among those who can, most contributions are still thin. As a counterpoint, 64% say AI supports innovation — felt value arriving before booked value, which fits how new technologies usually diffuse. A manager writing next year's budget should keep the two apart.

About 6%: what high performers do differently

Only about 6% of surveyed companies qualify as high performers. Set that against the agent data in the same survey — 62% of companies experimenting, no more than 10% scaled in any single function, covered in one year after the agent boom — and the divide between this small group and the rest becomes visible. From the public material, they did not simply buy more tools; they redesigned processes around AI: redefining roles, data flows and acceptance criteria, and treating compute, data governance and training as ongoing investment rather than one-off purchases.

How to read these numbers: two cautions

Before quoting any survey, know its edges. First, this is self-reported data: when a respondent says we use AI, no auditor checks the claim, so optimism bias is built in. Second, the definition of use is loose: one team member drafting emails with AI weekly and an AI auto-reply embedded in the service process can both count as regular use in a questionnaire. Numbers like these are best read for trends and relative gaps — 88% against last year's 78%, usage against scaling — not as industry benchmarks to match line by line.

A domestic footnote: Shanghai's supply-side data

China lacks a directly comparable adoption survey, but industry-side data offers context. Per Xinhua's report of April 30, 2026: in 2025, Shanghai's 394 above-scale AI enterprises — companies above a statistical revenue threshold — generated an industry scale of over RMB 637 billion, up 39.5% year on year; in the first quarter of 2026, the city's AI manufacturing output grew 19.2%. A supply side expanding at that pace means cheaper, better, more competitive tools. Whether the demand side can absorb them is the other question.

For smaller companies: most peers are also just starting

Read together, the numbers are oddly comforting for a smaller company: a scaling rate of about one third means most companies — including the loud ones — are still at trial and partial use. Entering now is not late. But the numbers also mark the fork in the road: the gap between buying tools and changing processes is the gap between 88% and about a third. A team with a limited budget does better to think through the six questions to ask before approving an AI project than to stockpile tools; and if tools were bought but sit unused, why employees end up not using them takes that problem apart.

What to watch when next year's numbers land

Usage has little headroom left — above 88%, further gains get increasingly cosmetic. Next year the comparisons that matter are three: whether the scaling rate moves visibly above one third; whether the distribution of EBIT impact thickens beyond mostly under 5%; and whether high performers exceed 6% of the sample. Usage rates measure the noise; these three measure the progress.