Companies abandoning most AI projects rose from 17 to 42 percent in a year

The share of companies abandoning most of their AI initiatives before production rose from 17 percent to 42 percent in a year, S&P Global Market Intelligence's 451 Research reported in 2025. The average organization in the survey scrapped 46 percent of its proofs of concept before production.

The data are older than the 2025 label suggests. The survey was fielded online from October 21 to November 25, 2024, among 1,006 mid-level and senior IT and line-of-business professionals in North America and Europe, and compared with 1,001 respondents surveyed in late 2023. The survey cannot say whether the rise continued after 2024.

The increase is far larger than the survey's margin of error of plus or minus 3 points. But it counts companies that abandoned most of their initiatives, not failed projects, and is often misquoted as 42 percent of AI projects failing. Staff resistance to AI was cited as a challenge by 28 percent of respondents, among the top five obstacles.

A proof of concept stopped because it failed a test set in advance is portfolio discipline.

What other studies add

MIT's Project NANDA reported in July 2025 that 95 percent of the organizations it studied were getting zero return on generative AI, despite an estimated $30 billion to $40 billion in enterprise investment. It found that 5 percent of custom enterprise AI tools reached production.

The report is labeled preliminary findings. Its evidence came from a review of more than 300 public AI initiatives, interviews with representatives of 52 organizations and 153 survey responses from senior leaders at four industry conferences, collected between January and June 2025. That is not a representative sample. The authors caution that the figures are indicative, drawn from interviews rather than official company reporting, and that definitions of success vary. Success meant marked, sustained productivity or P&L impact.

The study is often misquoted as showing that 95 percent of AI pilots fail. Its 5 percent figure refers to custom or vendor-sold enterprise tools reaching production. General-purpose tools such as ChatGPT and Copilot were widely piloted and deployed.

In McKinsey's 2026 State of AI survey, about 6 percent of respondents qualified as high performers, attributing 5 percent or more of EBIT to AI and describing its value as significant. Nearly three-quarters of them reported fundamentally redesigning workflows because of their AI use, against one-quarter of other respondents. That is an association within a small subgroup.

The three studies measure different things in different populations, and their figures should not be combined into a single failure rate.

What to count in your own portfolio

A rising abandonment rate is not necessarily bad news. A company that stops weak proofs of concept early may be showing discipline. The published S&P figures do not separate those cases from projects that simply stalled. A company's own records can.

• Count the AI proofs of concept stopped before production in the past year, as a share of those started. Compare it with S&P's 46 percent only loosely, since definitions differ.
• Classify each stop. A proof of concept stopped because it failed a test set in advance is portfolio discipline. One that stalled without a decision is a cost with nothing learned.
• Record the obstacles behind each stop, including staff resistance, which S&P's respondents placed among the top five.
• Before approving the next proof of concept, name the workflow it would change and the person accountable for redesigning it. In McKinsey's survey, high performers were far more likely to have redesigned workflows.