Skip to content
AI Tracker
← Back to Myths

Myth check: “MIT found 95% of AI pilots fail”

Myth check
Our verdictVerdict: Misleading

The 95% rests on 52 interviews and a 153-person survey with a strict test of success. It points to a real problem, not a measured rate.

The claim in full “An MIT study found that 95% of enterprise generative-AI pilots fail.”

52organisations interviewed for the report behind the “95%” figure

What the data shows

The claim spread in August 2025: an MIT study had found that 95% of companies’ generative-AI pilots fail. Many read it as proof that AI at work is mostly hype.

What’s true. The report, “The GenAI Divide: State of AI in Business 2025” from MIT’s NANDA project, does say “95% of organizations are getting zero return”, despite $30–40 billion invested. It describes a real pattern: custom AI tools built for companies often never make it into everyday use.

How it was measured. The figure rests on a review of 300+ public AI projects, interviews with people from 52 organisations and a survey of 153 senior leaders at four conferences. A tool only counted as a success if it had a clear, lasting effect on productivity or profit. The report is subtitled Preliminary Findings, and its authors say the numbers point in the right direction but come from interviews, not company accounts.

What the headlines skipped. Workers at over 90% of the surveyed companies regularly used personal AI tools for work, while only 40% of companies had bought an official subscription. Critics called the 95% figure statistically unsound. A separate Wharton survey of 800+ US business leaders (October 2025), measuring returns differently, found about three in four reporting positive returns.

What it means

Treat “95% of AI pilots fail” as a rough signal from a small, early study, not a measured failure rate. Many company AI projects do stall before they reach everyday use. But staff often use AI tools on their own, and a larger survey found most leaders reporting gains, so whether AI pays off depends on the company and the project.

Caveats

  • The report does describe a real problem: custom AI tools built for companies often fail to reach everyday use.
  • The Wharton survey uses a different design and measure, so it isn’t directly comparable.
Suggest a correction