From adoption to results: why EBIT is not moving
Almost nine in ten organizations use AI regularly. The share attributing enterprise-level financial impact to it has been frozen at 37% for a full year.
Executive summary
Usage generalized; results did not. Between 2025 and 2026 the share of organizations scaling AI across the enterprise rose from 38% to 44%, but the share attributing any EBIT impact stayed at 37%, the same as the year before.
The group that does get results is small and stable: about 6% of respondents. What separates it is not the model or the budget, but having redesigned workflows: 73% of them did, against 25% of everyone else.
Cost savings and revenue gains show up in different functions. Savings concentrate in software engineering and manufacturing; revenue improvements in marketing and sales, strategy and corporate finance.
The operational conclusion is uncomfortable: adding functions does not produce financial impact, redesigning one does. Horizontal expansion is easier to approve and easier to leave invisible in the income statement.
Context
For three years the metric that organized boardroom conversations was adoption: how many functions covered, how many active users, how many pilots open. That metric is now saturated and no longer discriminates between organizations.
The metric that does discriminate is more demanding and slower: whether AI usage shows up in operating results. There the picture has not moved for a full year, while the associated spending keeps growing.
Figure 1 · What moved and what did not, between 2025 and 2026
% of surveyed organizations, 2026Table 1 · Where savings appear and where revenue does
% of organizations reporting each effect| Reported effect | Function | Organizations |
|---|---|---|
| Cost savings | Software engineering and manufacturing | 56% |
| Revenue gains | Marketing and sales | 67% |
| Revenue gains | Strategy and corporate finance | 65% |
| Revenue gains | Product or service development | 62% |
| Improved innovation | Across the organization | 64% |
| Worsened cost metrics | Any function | up to 7% |
Business implications
Pick one function and redesign the whole flow. The high-performer group is distinguished by this and little else. It is the variable with the greatest explanatory power over financial impact.
Stop counting covered functions. It is the metric that rises most easily and correlates least with operating results. It serves the dashboard, not the decision.
Put the metric in the income statement, not the usage dashboard. Cost per transaction or cycle time are verifiable; "active users" does not answer to an audit committee.
Methodology & data
Two editions of the same McKinsey survey are contrasted: the 2025 one (n = 1,993 across 105 countries), republished in Stanford HAI's AI Index 2026, and the 2026 one (n = 1,719 across 97 countries, fielded between 4 May and 8 June). Both collect voluntary executive responses, not a census of firms.
The 2026 edition measures "AI" without separating generative AI: the series are not strictly comparable across years.
References
McKinsey & Company (25 August 2026). The state of AI in 2026: On the road to ROI. · Stanford HAI (2026). The AI Index 2026 Annual Report, chapter 4.