Primary sources. Methodology shown.
Why AI projects fail, and what AI implementation really costs
The most quoted statistic in this field is that 80.3% of AI projects fail. No study says that. RAND wrote “by some estimates, more than 80 percent”, and the decimal was added somewhere downstream.
We checked every figure on this site against the document it came from. Of 15 sources, 9 are routinely misquoted in ways we could verify against the original, including one breakdown, widely credited to RAND, that RAND never published.
No sign-up. Nothing stored. Nothing to buy.
80.3%
What RAND actually wrote
“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.”
RAND, 2024. Punctuation normalised, wording unchanged.
01 · Sources, not summaries
Every number carries its scaffolding
Sample size, method, fieldwork window, and a link, pinned in the margin beside the claim. Where a study publishes no sample, we say so instead of inventing one.
02 · Predictions are labelled
A forecast is not a measurement
Two of the most repeated AI failure figures are Gartner forecasts about years that had not happened yet when they were written. They get an amber dot, not a headline.
03 · Nothing to sell
No consultancy, no gate, no call
We have run no surveys and delivered no AI projects. Our authority is that we read the primary research and show our working.
Start here
The four questions people actually ask
What percentage of AI projects fail?
The honest answer needs four studies reconciled, because each measures something different. One of the four numbers you have seen does not exist.
Why do they fail?
RAND puts the causes in leadership and organisation, not technology. It attaches no percentage, whatever you have read.
What does it cost?
Nobody credible publishes a figure. The ranges you have seen are agency marketing. Here is what is actually documented.
Will mine fail?
No tool can tell you that, and any that claims to is lying. This one checks your project against the factors the studies named.
Answers
Questions, answered from the sources
What percentage of AI projects fail?
RAND reported in 2024 that, by some estimates, more than 80% of AI projects fail, about twice the rate of non-AI IT projects. It publishes no more precise figure, and the widely repeated "80.3%" appears in no study. MIT reported separately in 2025 that 95% of organisations were getting zero return from generative AI. The two measure different things.
Why do most AI projects fail?
RAND interviewed 65 data scientists and engineers and concluded the root causes are overwhelmingly leadership and organisational rather than technical. Gartner names poor data quality, inadequate risk controls, escalating costs and unclear business value in its forecasts of abandonment.
Can I check whether my own project is at risk?
Yes. This site publishes a free diagnostic that scores a project against the factors named in the published studies. It is not a validated statistical model, it stores nothing, and it requires no sign-up.