Skip to content
back to the archive page
#AI

McKinsey State of AI 2026: scaling is rising, but the EBIT effect is not (Rubric #AI)

#AI #Agents #Research #Metrics #Management #Economics

On August 25, McKinsey published its new “State of AI 2026.” It has a remarkably precise subtitle: “On the road to ROI.” The data suggest that it is no longer enough to “adopt” AI by giving people access to a tool. Organizations now have to prove that local acceleration survives the difficult journey through workflows, budgets, and the P&L.

I previously assembled the earlier editions of the series into a time series as part of T-Technologies’ AI4SDLC research. The share of respondents reporting AI use in at least one function rose from 55% in 2023 to 72% in early 2024, 78% in mid-2024, and 88% in 2025. The new survey puts it at 89% :)

McKinsey Exhibit 1 chart on AI adoption and scaling from 2017 to 2026

This is a useful trend, but not a perfectly clean time series: before 2025, McKinsey asked about “adoption” while leaving the term undefined, then switched to “regular use.” Each edition is also a new cross-sectional snapshot rather than a study of the same panel of companies. Depth of use has therefore become more informative than penetration alone.

There is movement on that front:

  • Among respondents at organizations already using AI, 44% report enterprise-wide scaling, up from 38% a year earlier;
  • 56% report AI use in three or more functions, up from 51%;
  • At organizations with at least $1 billion in revenue, the share scaling AI agents in at least one function rose from 27% to 40%. At smaller organizations, it was essentially flat at 22%.

McKinsey Exhibit 2 chart comparing AI-agent scaling in large and smaller organizations

The financial picture, however, has barely moved. 80% of respondents who personally use AI say it has improved their productivity, and half say it helps them make better decisions.

McKinsey Exhibit 5 chart on perceived AI effects on productivity, skills, and decisions by role

Yet only 37% of respondents at organizations that regularly use AI attribute any positive EBIT contribution to it. A year ago, the figure was 39%.

The same “AI high performers” — organizations whose respondents attribute at least 5% of EBIT to AI and report “significant” value — still make up about 6% of the sample, or 92 respondents. Nearly three-quarters of them, however, have fundamentally redesigned workflows because of AI; last year the figure was 55%, while today it is only one-quarter among other organizations. This is a correlation in a self-reported survey, not proof of causation. Still, the direction is familiar: placing a tool on top of an old process produces activity faster than it produces a new economic outcome.

The report also contains two signs of more mature operations

  1. Among respondents at organizations using AI, 20% say operating costs, including tokens, constrained use of the technology
  2. 32% say their organization decided against purchasing at least one software product or feature because it could build the required functionality in-house with coding agents Meanwhile, 60% expect AI investment to keep growing. That is not yet a forecast of a SaaS collapse, but it is a notable signal for the “build versus buy” discussion.

There is also a useful reality check on workforce forecasts. Among respondents at organizations using AI, in 2025, 32% expected a decline in total head count. A year later, only 14% said such a decline had actually happened. Now 39% expect one over the coming year. McKinsey repeated the comparison for the 552 people who answered both surveys and found the same pattern.

The survey ran from May 4 to June 8, 2026: 1 719 respondents in 97 countries, with responses weighted by each country’s share of global GDP. These are self-reports, not audited financial statements, so “any EBIT impact” should not be read as a measured ROI magnitude.

Over four years, the series’ central question has shifted from “Is AI used anywhere?” to “Does individual productivity travel through a redesigned workflow into enterprise economics — and what does that journey cost?” The measurement chain should now look like this: “individual impact → workflow → function economics → EBIT,” with cost and risk tracked alongside it. Otherwise, the road to ROI ends on a dashboard where adoption is already green while the money is still gray :)

#AI #Agents #Research #Metrics #Management #Economics