ai project failure statistics
The numbers everyone quotes about AI failure, traced back to where they came from — what each one actually measures, and how it gets misquoted.
How many AI projects actually make it.
Measured figures — surveys of what organizations report actually happened. They use different denominators, so read each one on its own terms.
of AI projects make it into production, on average — and it takes about 8 months to go from prototype to production.
A respondent-reported average, not a count of projects. The top barrier respondents named was estimating and demonstrating value (49%) — not data.
of companies abandoned most of their AI initiatives in 2025 — up from 17% the year before. The average organization scrapped 46% of its proofs of concept before production.
Companies abandoning most of their initiatives — not 42% of projects, and not “at least one”.
AI proofs of concept reached production. The other 88% never made it to widescale deployment.
IDC attributes it to “the low level of organizational readiness in terms of data, processes and IT infrastructure”. No public IDC page — the figure is cited via CIO.com.
of generative AI projects were abandoned after proof of concept, by Gartner’s own later count — against the 30% it had forecast in 2024.
The rare case where a forecast was checked against what happened. Reality came in worse.
The two numbers everyone quotes.
Both are real. Neither says what the headline says.
of AI projects fail “by some estimates” — twice the failure rate of IT projects that don’t involve AI.
RAND is repeating other people’s estimates, not reporting a count. Its own finding is five root causes — the first is misunderstanding the problem; the second is lacking the data needed to train an effective model.
Not: “RAND found 80% of AI projects fail.”
of organizations are getting zero return from generative AI, despite $30–40 billion of enterprise investment.
About measurable P&L impact from GenAI pilots — not project failure, and not AI in general. The authors blame a “learning gap” (tools that don’t adapt to how work is done), not model quality. The report has no stable mit.edu page; it circulates as a PDF.
Not: “95% of AI projects fail.”
Five misquotes, corrected.
“95% of AI projects fail.”
95% of organizations saw no measurable P&L return from generative AI within about six months, in a preliminary MIT NANDA study.
“RAND found that 80% of AI projects fail.”
RAND cites estimates of more than 80%. Its own research is 65 interviews about root causes.
“Gartner says 60% of AI projects fail.”
Gartner forecasts that 60% of AI projects without AI-ready data will be abandoned through 2026.
“42% of AI projects are abandoned.”
42% of companies abandoned most of their AI initiatives in 2025 (S&P Global).
“88% and 95% — the failure rate is around 90%.”
IDC’s 88% measures reaching production; MIT’s 95% measures P&L return. Different questions — never average them.
Forecasts, not results.
Predictions by an analyst firm. Useful as a signal of where analysts see risk — not evidence of what happened.
of generative AI projects will be abandoned after proof of concept by the end of 2025 — due to poor data quality, inadequate risk controls, escalating costs or unclear business value.
Poor data quality is the first reason named. Gartner’s later count put the outcome at 50% (above).
of AI projects unsupported by AI-ready data will be abandoned through 2026.
Conditional: it is about projects without AI-ready data, not all AI projects.
Not: “Gartner found 60% of AI projects fail.”
of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls.
This one does not name data as a cause. Don’t recruit it for a data argument.
Where the data shows up.
Studies that ask about data directly. This is the first mile, measured.
of organizations either do not have, or are unsure if they have, the right data management practices for AI.
Published alongside the 60% forecast above. Read the release for the sample size; secondary sources disagree on it.
of data professionals say their data is of sufficient quality and accessibility for effective AI. 64% name data quality as their top data-integrity challenge, up from 50% a year earlier.
A vendor-sponsored survey — but run with a university business school, and its question is exactly the first-mile one.
of chief data officers name “data quality, completeness and readiness” among the top obstacles to getting generative AI into production.
Vendor-sponsored. Some trade press prints 42%; Informatica’s own release says 43%.
more, as a share of revenue, is what organizations with successful AI initiatives invest in data quality, governance, AI-ready people and change management.
“Up to” four times. A correlation between foundations spending and success — not proof that one causes the other.
Value, at scale, is still rare.
Broad studies of whether AI is paying off — the outcome every failure statistic is a proxy for.
of organizations attribute any EBIT impact to AI — flat on the year before. AI high performers remain about 6%.
“Any” EBIT impact — for most, a small share of the total.
of companies are “future-built” and achieving value from AI at scale; 60% report minimal or no material value.
BCG’s 2024 edition put the share yet to show tangible value at 74%.
What the statistics agree on
Different methods, one pattern: most AI stalls between the pilot and the payoff.
Where studies look for causes, data readiness is named again and again — by Gartner, RAND, IDC, Informatica and Precisely — next to unclear value, cost and risk. Not every study blames data: Gartner’s production survey points at proving value first, and MIT points at a learning gap. Honest reading: data is not the only reason AI fails. It is the reason you can check before you start.
That is the first-mile problem in one line — and the reason the AI-ready data checklist exists.
Method.
Every figure was traced to the publisher’s own release or report page. Where no public primary page exists, the outlet that first reported it is named instead, and the card says so. Forecasts are labeled as forecasts. Sample sizes appear only where the publisher states them. Vendor-sponsored surveys are marked as such.
Last checked . The page is reviewed every quarter, and when a forecast’s deadline passes, the outcome is added next to it.
AI failure, answered.
What percentage of AI projects fail?
There is no single honest number, because studies count different things. Measured figures cluster around half: Gartner found that only 48% of AI projects make it into production; S&P Global found that 42% of companies abandoned most of their AI initiatives in 2025; IDC found that 4 in 33 proofs of concept reached production. The widely quoted 80% (RAND) is a cited estimate, and the 95% (MIT NANDA) is about returns from generative AI — not project failure.
Is it true that 95% of AI projects fail?
No. The 95% comes from MIT NANDA’s preliminary 2025 report, The GenAI Divide, which found that 95% of organizations were getting zero measurable P&L return from generative AI within about six months. It measures financial return from GenAI pilots, not whether AI projects fail, and the authors attribute it to a learning gap rather than to the models.
Why do AI projects fail?
Sources name several causes: unclear business value, cost, risk controls — and data. RAND’s interviews found five root causes, including misunderstanding the problem and lacking the data needed to train an effective model. Gartner forecasts that organizations will abandon 60% of AI projects unsupported by AI-ready data through 2026, and IDC attributes failed proofs of concept to low readiness in data, processes and IT infrastructure.
What percentage of AI projects make it to production?
In Gartner’s 2024 survey of organizations that had deployed AI, an average of 48% of AI projects made it into production, taking about eight months from prototype. IDC’s research with Lenovo found that only 4 of every 33 AI proofs of concept reached production.