The RAND study on why AI projects fail

What RAND actually did

RAND Corporation published RRA2680-1 in 2024. The method was 65 semi-structured interviews, 50 with industry professionals and 15 with academic researchers, conducted between August and December 2023. That is the whole dataset: what a set of practitioners and researchers said in conversation, not a count of projects, an analysis of postmortems, or a statistical model of failure.

That method matters for how much weight the figures can carry. 65 structured conversations can surface real, recurring patterns, which is what the five root causes are. They cannot produce a precise industry-wide failure percentage, and they cannot produce a quantified breakdown of how AI projects end. Anyone citing RAND for either of those two things is citing a method the report does not have.

The headline figure, in RAND's own words

RAND's own sentence: "by some estimates, more than 80 percent of AI projects fail, twice the rate of failure for information technology projects that do not involve AI." Read the hedge. RAND is reporting what "some estimates" say, not a figure it calculated from its own interviews. The 65 conversations back the qualitative pattern of causes; they are not the source of the 80 percent number.

This is the only place in the report where a failure percentage appears. There is no decimal point, no second figure, and no breakdown of what "fail" means beyond the general sense used throughout the report: a project that did not deliver on what it set out to do.

The 84% figure, and the leadership finding

RAND's most consequential sentence: "More than any other type of issue, our interviewees noted that failures driven by the decisions and expectations of the organization's business leadership were far and away the most frequent causes of project failure. Eighty-four percent of our interviewees cited one or more of these root causes as the primary reason that AI projects would fail."

That is 84% of 65 people, describing which cause they saw most often. RAND breaks the leadership-driven category into four sub-causes: optimising for the wrong business problem, using AI to solve problems that were already simple to solve, overconfidence in what AI can do, and underestimating the time a project takes. All four are decisions made by the people managing the project, before or during the build, not failures of the model itself.

Academia and industry fail for different reasons

RAND interviewed both groups and did not treat their findings as one. On academic projects, it writes: "In our interviews, we learned that when AI projects fail, they do so because of a misalignment in incentives rather than an overall technical barrier to product delivery. In short, overcoming an AI failure is more about humans than the machines." On industry projects, its closest equivalent is narrower: "Overall, interviewees expressed that the most common root cause of failure was the business leadership of the organization misunderstanding how to set the project on a pathway to success."

Both quotes point at people rather than models, and both tend to get flattened, in retelling, into the same generic line about AI failure being a people problem rather than a technology problem. That collapses two distinct findings, from two distinct populations, into one. Use the industry line for industry claims and the academic line for academic ones. RAND does not use them interchangeably, and neither should anyone quoting it.

What RAND recommends

Five principles for industry, in RAND's wording:

• Ensure that technical staff understand the project purpose and domain context
• Choose enduring problems
• Focus on the problem, not the technology
• Invest in infrastructure
• Understand AI's limitations

For academic teams, two more: overcome data-collection barriers through partnerships with government, and expand doctoral programs in data science for practitioners.

What RAND did not say

Three specific claims circulate under RAND's name. We checked all three against the primary text. None of them is there.

"80.3% of AI projects fail"

This is the figure you have most likely seen, often presented with the confidence of a measured statistic. RAND's actual wording is "by some estimates, more than 80 percent," a hedge on someone else's estimate, with no decimal place and no single named source for the "some estimates". 80.3% appears nowhere in RRA2680-1.

The four-way outcome breakdown

A tidy split often attributed to RAND: 33.8% of projects abandoned, 28.4% delivering no value, 18.1% partially successful, and 19.7% succeeding. It appears nowhere in the report, and RAND's method could not have produced it: 65 semi-structured interviews yield themes and quotes, not a quantified taxonomy of outcomes across an industry. The four numbers sum to exactly 100.0%, which is what fabricated data tends to do, and which real survey data, with its rounding and non-response, rarely does so neatly.

The 84% conflation

The 84% you will see quoted as "84% of AI project failures are caused by leadership" is real, but narrower: it is the share of RAND's 65 interviewees who named a leadership-driven cause as the primary one. It describes an opinion split among a small expert sample, not an audit of the industry. To make it more confusing, a second, unrelated 84% sits in the same report, taken from Cisco's AI Readiness Index: "84 percent of business leaders responded that they believe that AI will have a significant impact on their business, and 97 percent of business leaders reported that the urgency to deploy AI-powered technologies has increased." Two different 84% figures, describing two different things, in one document.

Questions, answered from the primary text

Related: the practitioner summary of why AI projects fail, and MIT's report on the generative AI learning gap, which gets the same source-checking treatment.