AI in the Workplace Statistics (2026)

McKinsey's November 2025 survey found almost all respondents saying their organisation uses AI in at least one function — which makes the headline adoption number close to useless for planning. The useful figure sits underneath it: nearly two-thirds say they have not begun scaling AI across the enterprise.

Practically, that means most AI in the workplace today lives in a department, a tool subscription or a pilot with an owner and no mandate. Statistics that describe a company as an AI adopter are describing that state, not a transformed operating model.

Respondents readily report cost and revenue benefits at use-case level, and 64% credit AI with enabling innovation. Only 39% report EBIT impact at enterprise level. That combination is exactly what a portfolio of small, uncoordinated wins looks like on a P&L: visible locally, invisible in aggregate.

The commonly cited reasons are unglamorous — the work stays organised around the pre-AI process, so the saved minutes never turn into fewer steps, shorter cycle times or reduced spend.

62% of respondents say their organisations are at least experimenting with agents. Experimentation is a low bar, and it should be read as such in any chart that shows a steep agent-adoption curve.

The distinction worth tracking through 2026 is whether agents are handed a business process with an owner, a service level and an audit trail, or whether they remain assistants sitting beside a human who still does the work.

80% of organisations set efficiency as an objective. The companies capturing the most value tend to add growth or innovation objectives on top, and half of the AI high performers intend to use AI to transform the business rather than trim it. Most of them are redesigning workflows.

Microsoft's Work Trend Index frames the same shift from the org-design side, describing an emerging model in which human teams direct AI capacity rather than simply consume it. Read it as a direction of travel, not as measured outcomes.

Asked about workforce size over the coming year, 32% expect decreases, 43% expect no change and 13% expect increases. Any brief that reports only the first number is describing a third of the sample as if it were all of it.

For workforce planning, the honest reading is that role content is changing faster than role counts, and that the survey data does not yet support a single confident headcount trajectory.

Most workplace AI statistics measure access or trial: whether an organisation has bought a tool, or whether an employee has ever used one. Sustained weekly use by the people whose work the tool was meant to change is a much smaller number, and it is the one that correlates with any measurable output.

When comparing surveys, read the question wording. "Have you used AI at work?" and "Do you use AI in your core tasks most weeks?" produce answers that can differ by a factor of three within the same population.

Self-reported savings cluster in drafting, summarising and searching — tasks with a clear before-and-after and no formal cycle time. They are real, but they are also the easiest to overstate, because the counterfactual is a draft nobody timed.

Savings that survive scrutiny tend to appear where a process already had a clock on it: ticket handling, first-response times, code review turnaround. If you plan to publish an internal figure, pick a task with an existing baseline before rollout rather than after.

In most surveyed organisations, employee use runs ahead of written policy. That gap is where the governance risk sits: not in the tools that were procured and reviewed, but in the free consumer accounts used to finish something on a deadline.

A short, specific policy — what may be pasted, into which tools, and what must be reviewed by a person — closes more exposure than a long framework nobody reads. It also makes usage data honest, because staff no longer have a reason to hide the tools they actually use.