AI-first company

Claim check

“95% of generative AI pilots at companies are failing (MIT)”

The figure is in the report, but the report is preliminary and not representative. It rests on 52 interviews, 153 survey answers collected at conferences and a review of public announcements. Its authors call the figures “directionally accurate” rather than based on company reporting, and it defines success in two different ways.

By
Last reviewed
Corrections
Report an error
Verdict Needs correcting

The source exists, but the claim gets the year, the population or the wording wrong.

Kind of figure Self-reported

What people said in a survey or interview, not measured.

The claim

95% of generative AI pilots at companies are failing, according to MIT.

What the primary source says

The report is The GenAI Divide: State of AI in Business 2025, dated July 2025, from Project NANDA at the MIT Media Lab. Its notes page calls it “Preliminary Findings from AI Implementation Research from Project NANDA”.

How it was done. The authors reviewed over 300 AI initiatives that companies had made public. They interviewed people from 52 organizations. They collected 153 survey answers from senior leaders at four industry conferences. The research ran from January to June 2025.

Where the 95% appears. The summary says “95% of organizations are getting zero return”. A later section speaks of “the 95% failure rate for enterprise AI solutions”.

What the figure stands on. The report gives a funnel for AI tools built for a specific task. 60% of organizations evaluated such tools, 20% reached the pilot stage and 5% reached production. All three figures are shares of all organizations, not of pilots.

What it says about itself. Beside the funnel, the authors write: “These figures are directionally accurate based on individual interviews rather than official company reporting.” Its list of limits says that the sample “may not fully represent all enterprise segments or geographic regions”, and that organizations willing to talk may differ from those that declined.

What is wrong with the claim

  • It is not a count of pilots. The 5% is a share of organizations that got a tool into production. The report does not say that 95% of pilots failed.
  • Success is defined in two ways. One note counts a tool as successfully implemented when “users or executives have remarked” on a lasting effect on productivity or profit. The methods section defines success as “deployment beyond pilot phase with measurable KPIs”, meaning set performance measures.
  • The sample is small and chosen, not random. 52 interviews and 153 answers collected at conferences cannot stand for “companies” in general.
  • The figures come from what people said. The authors say so themselves: interviews, not company records.
  • It is preliminary. The report labels its own findings preliminary.

The report can be cited as an early, small study that found few task-specific AI tools reaching production. It is not evidence that 95% of AI projects fail. For measured results from AI at work, see the studies on this site.

What to cite instead

A preliminary 2025 report from MIT’s Project NANDA, based on interviews at 52 organizations, 153 survey answers from senior leaders at industry conferences and a review of over 300 public AI initiatives, estimated that 5% of organizations had task-specific generative AI tools in production. The authors describe their figures as “directionally accurate based on individual interviews rather than official company reporting”.

Sources

  1. Challapally, Pease, Raskar and Chari (July 2025). The GenAI Divide: State of AI in Business 2025. MIT NANDA, preliminary findings. Copy of the report hosted by MLQ.ai. mlq.ai
  2. Fortune (18 August 2025). MIT report: 95% of generative AI pilots at companies are failing. The press coverage behind the claim. fortune.com