Summary: McKinsey’s November 2025 survey finds 88 percent of organizations regularly using AI and 62 percent at least experimenting with AI agents, yet only 39 percent report any enterprise EBIT impact, usually under 5 percent of EBIT. Cycle-time dashboards can go green while the P&L stays flat. The same pattern shows up when enterprise AI projects blow budgets without attributable earnings.
Use is common. Enterprise EBIT is rare.
The State of AI survey, published November 5, 2025, reports that 88 percent of respondents’ organizations regularly use AI in at least one business function, up from 78 percent a year earlier.
The earnings line is far smaller. The same survey reports that 39 percent of respondents attribute any level of EBIT impact to AI, and most of those say that less than 5 percent of their organization’s EBIT is attributable to AI use. On McKinsey’s reading, AI use has not yet significantly affected enterprise-wide EBIT for most organizations.
The report also states: “Nearly two-thirds of respondents say their organizations have not yet begun scaling AI across the enterprise.” Approximately one-third have begun to scale, and the majority remain in the experimenting or piloting stages.
Agents sit in the experiment layer, so the P&L never sees them.
The agent layer is thinner than the general-use layer. Sixty-two percent of survey respondents say their organizations are at least experimenting with AI agents. That figure is the on-page sum of 23 percent scaling an agentic AI system somewhere in the enterprise and an additional 39 percent who have begun experimenting.
Even the organizations scaling agents mostly confine them to one or two functions. In any given business function, no more than 10 percent of respondents say their organizations are scaling AI agents. An experiment that never leaves one team cannot move enterprise EBIT. A pilot that succeeded locally and failed in production has the same shape: the green local result never arrives at the line that EBIT measures.
Use-case savings can be real and still miss the enterprise line.
Enterprise-wide bottom-line impact remains rare even as respondents report cost benefits from individual use cases, most often in software engineering, manufacturing, and IT. Those local benefits can be real and still sit below the enterprise line.
Eighty percent of respondents say their companies set efficiency as an objective of their AI initiatives. A cycle-time dashboard can go green while the EBIT line does not move because savings are absorbed into later work, reinvested in more experiments, left unallocated, or remain too small to clear materiality. Cycle-time wins can also be erased by unit costs before they reach earnings, a pattern already visible in how teams budget for AI agent workflows.
The programs that move EBIT redesign the workflow, then finance can name the value.
Respondents who attribute EBIT impact of 5 percent or more to AI and who say their organization has seen significant value are McKinsey’s AI high performers, about 6 percent of respondents. They are nearly three times as likely as others to say their organizations have fundamentally redesigned individual workflows.
The report is direct: this intentional redesigning of workflows has one of the strongest contributions to achieving meaningful business impact of all the factors tested. The P&L moves when the work itself changes and finance can attribute the result.
Finance is still struggling to name that result. The FinOps Foundation 2026 survey lists “Determining AI value/ROI” among the visibility and value challenges of applying FinOps to AI, because investments are often exploratory and returns are hard to define early. A practitioner on the same page puts it plainly: “Is your AI providing value? No one can answer that question yet.” Ninety-eight percent now manage AI spend, up from 31 percent two years ago. Managing the bill does not, by itself, put a number on the enterprise line.
