The claim
More than 80% of AI projects fail, twice the rate of IT projects without AI, according to a RAND study.
How the claim travelled
Each step is a publication that passed the claim on, oldest first, with its exact words and what changed.
-
Fortune (Jeremy Kahn) · archived copy
“That’s borne out in a slew of recent surveys, where business leaders have put the failure rate of A.I. projects at between 83% and 92%.”
What changed: Dead end. No survey is named or linked. The figure is 83% to 92%, and it is about financial returns. The article is built around the head of an AI software company.
-
Harvard Business Review (Iavor Bojinov) · archived copy
“Most AI projects fail. Some estimates place the failure rate as high as 80%—almost double the rate of corporate IT project failures from a decade ago.”
What changed: Dead end. No source is given for the 80% or for the comparison. The 80% is an upper limit (“as high as”), and the comparison is with IT projects ten years earlier.
-
RAND Corporation (Ryseff, De Bruhl and Newberry) · archived copy
“By some estimates, more than 80 percent of AI projects fail. This is twice the already-high rate of failure in corporate information technology (IT) projects that do not involve AI.”
What changed: RAND quotes the two articles above: note 13 (Fortune) for the 80%, note 14 (Harvard Business Review) for the comparison. Fortune’s “between 83% and 92%” becomes “more than 80 percent”. “Almost double” the rate of “a decade ago” becomes “twice” today’s rate. RAND’s own research was 65 interviews; it counted no projects.
-
“New research has claimed the vast majority (80%) of AI-based projects fail, double the normal failure rate for non-AI tech proposals.”
What changed: “By some estimates” is dropped. The figure becomes a claim of “new research”, and the next sentence names the RAND study.
-
“80% of AI projects fail to deliver their intended business value. That's twice the failure rate of regular IT projects, and it has barely moved in three years, according to RAND Corporation's analysis of 2,400+ enterprise AI initiatives.”
What changed: A method is invented. RAND did not analyse 2,400 AI initiatives, and the article dates the figure 2025. “Fail” becomes “fail to deliver their intended business value”.
What the primary source says
The report is The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, published by RAND in August 2024. RAND is a research institute. The work was paid for from research funds of its National Defense Research Institute, which US defence bodies sponsor. RAND has no commercial interest in the figure.
How it was done. The authors interviewed 65 data scientists and engineers with at least five years of experience building AI models. 50 worked in industry and were found through LinkedIn; 15 were academics. The interviews took place between August and December 2023. The authors counted a project as failed when the organization saw it as a failure. Projects that only used ready-made large language models, the kind behind ChatGPT, were left out.
What it found. The interviews gave five root causes of failure. The first: people in the business “misunderstand—or miscommunicate—what problem needs to be solved using AI”.
Where the 80% appears. The introduction says: “By some estimates, more than 80 percent of AI projects fail. This is twice the already-high rate of failure in corporate information technology (IT) projects that do not involve AI.” The summary repeats it. One note cites an article in Fortune for the 80%. Another cites an article in Harvard Business Review for the comparison.
Where the trail ends. The Fortune article says that “a slew of recent surveys” put the failure rate at 83% to 92%, and names none of them. The Harvard Business Review article says that “some estimates” put it “as high as 80%”, and gives no source. The comparison with IT projects also comes from that article, with no source.
What is wrong with the claim
- RAND did not measure it. RAND counted no projects. It quoted the figure from two articles, with “By some estimates”.
- There is no original. Both articles name no survey, no number of projects and no meaning of “fail”.
- The comparison grew on the way. RAND cites the Harvard Business Review article for it. That article says “almost double” the rate of IT projects “a decade ago”. RAND says “twice” the rate of IT projects today.
- Later repeats add details that are not in any source. One vendor’s article says RAND analysed “2,400+ enterprise AI initiatives”. RAND did not.
- It is often applied to generative AI. RAND’s own interviews left out projects that only used ready-made language models, so the report says nothing about projects such as a chatbot built on ChatGPT.
What other sources found
This check found no source that counted the failure rate of AI projects in general. Surveys ask managers how many of their AI projects went into use, and the answers differ from 80%. In a survey by Gartner, which sells research and advice to IT leaders, 644 respondents in the US, Germany and the UK said in late 2023 that “on average, only 48% of AI projects make it into production”. That is what respondents told the survey, not a count of projects.
RAND’s report is worth citing for its five root causes, from 65 interviews. It is not a source for the share of AI projects that fail. For measured results from AI at work, see the studies on this site.
What to cite instead
RAND’s 2024 report on why AI projects fail did not measure a failure rate. It interviewed 65 experienced data scientists and engineers between August and December 2023, and named five root causes, the first being that people misunderstand or miscommunicate what problem the AI should solve. The “more than 80 percent” figure in its introduction rests on a Fortune article that names no source for it.
Sources
- Ryseff, De Bruhl and Newberry (2024). The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed: Avoiding the Anti-Patterns of AI. RAND Corporation, Research Report RR-A2680-1. rand.org
- Kahn (26 July 2022). Want your company’s A.I. project to succeed? Don’t hand it to the data scientists, says this CEO. Fortune. RAND’s note 13. fortune.com
- Bojinov (November–December 2023). Keep Your AI Projects on Track. Harvard Business Review. RAND’s note 14. hbr.org
- Gartner (7 May 2024). Gartner Survey Finds Generative AI Is Now the Most Frequently Deployed AI Solution in Organizations. Press release. gartner.com