Checked AI agent stats reveal adoption figures vary widely by what is counted

A source-checked analysis from bdautomated finds that the most frequently cited AI agent statistics measure intention, and actual use within any single department does not exceed 10 percent.

Seventy-five figures taken from 18 publishers have been traced back to the original reports, with details of who was surveyed and what was counted. The dataset is freely available for download.

On Tuesday, bdautomated released “AI agent statistics 2026: every number checked at its source”, a free reference page and dataset that traces 75 widely quoted statistics on AI agents and business AI use to the original documents.

The analysis shows that the figures conflict because they measure different things. McKinsey’s 2025 global survey found that 62 percent of organisations were at least experimenting with AI agents, 23 percent had scaled one somewhere in the company, and in any single business function, no more than 10 percent had done so.

Surveys that record any adoption or intent report much higher numbers: 79 percent of U.S. executives told PwC in April 2025 that agents were already being adopted at their companies, while a Capgemini survey that rechecked what respondents meant by “agent” found 14 percent had implemented one. Across all U.S. businesses of every size, the Census Bureau found 19.8 percent using AI in any business function as of May 2026.

The page also explains the figure that shook markets in 2025. MIT Project NANDA’s finding that “95 percent of organisations are getting zero return” measured profit-and-loss impact within roughly six months of a pilot, in a sample of 52 interviews, 153 conference survey responses and 300 public deployments, and its authors call the findings preliminary. It does not mean that 95 percent of AI projects fail. Gartner’s prediction that over 40 percent of agentic AI projects will be cancelled by the end of 2027 is a June 2025 forecast; no cancellations have been counted yet.

Every figure on the page passed four checks: the number appears in the original document; the exact place and a verbatim quote are recorded; what it measures is written in plain words, including who was asked, how many and when; and it is compared against other sources, with disagreements shown rather than averaged. Market-size forecasts were excluded because the reports behind them are paid and cannot be verified. The dataset is published as CSV and JSON under a CC BY 4.0 licence, and the page includes a corrections address.

“Two headlines in the same week said almost nobody has AI agents running and almost everybody does, and both were quoting real surveys. We wanted the page we could not find: what each survey actually asked, so a business owner can tell which number is about a company like theirs,” said a spokesperson for bdautomated.

The page contains four charts that other publications may embed with attribution, and a table of all 75 figures with their sources, dates, samples and quotes.

Read the analysis: https://bdautomated.com/ai-agent-statistics/

Download the data: https://bdautomated.com/data/ai-agent-statistics.csv