AI project failure rate
What is the AI project failure rate?
The headline numbers are stark and they mostly agree. Across the major studies, the large majority of enterprise AI projects never deliver measurable value, and the cause is rarely the technology.
The short answer
Most research puts the AI project failure rate between 80% and 95%. RAND finds more than 80% of AI projects fail, about twice the rate of non-AI IT projects, and MIT’s 2025 study finds 95% of enterprise generative-AI pilots produce no measurable P&L impact. The failures are overwhelmingly about how projects are chosen, funded, and governed, not about model quality.
- RAND: more than 80% of AI projects fail, roughly twice the rate of non-AI IT projects.
- MIT NANDA: 95% of enterprise generative-AI pilots produce no measurable P&L impact.
- McKinsey: 88% of organisations use AI, but only 39% see any enterprise EBIT impact.
- Gartner: over 40% of agentic AI projects will be canceled by the end of 2027.
Why the numbers vary from 80% to 95%
Different studies measure different things, which is why the failure rate is quoted as a range rather than a single figure. RAND’s 80% comes from interviews about AI projects that never reach reliable production. MIT’s 95% measures a narrower, harsher bar: enterprise generative-AI pilots that produce no measurable impact on the P&L.
They point the same direction. Whether you define failure as “never shipped” or “shipped but moved no money,” the large majority of AI initiatives do not pay off. McKinsey’s finding that only 39% of adopters see any enterprise EBIT impact, despite 88% using AI, is the same story from the value side.
Why AI projects fail
The consistent finding across RAND and the others is that failure is a governance and prioritisation problem, not a technology one. The most common root causes are organisational:
- Misunderstood problem: the project solves for the wrong metric or does not fit the workflow.
- No evidence before capital: the business case rests on assumptions no one tested.
- Weak data foundations: the model has no reliable, relevant data to learn from.
- Technology-first thinking: teams chase the newest model instead of the real problem.
- No kill discipline: losing bets keep their funding long past the point the evidence turned.
The failure rate is a prioritisation signal, not a reason to stop
An 80-to-95% failure rate does not mean AI does not work. It means most bets, judged one at a time, were never the ones worth funding. The teams that beat the average do not have better luck; they compare every bet up front and demand evidence before capital.
That reframes the number. A high failure rate is exactly why a portfolio approach pays off: if most bets fail, the value is in cheaply identifying the few that will not, before the build budget is committed.
How to beat the average
The move that changes the odds is making evidence cheap. Pretotyping, created at Google and taught at Stanford, produces real behavioural signal on a bet in days, for a fraction of a build. Score every bet, pretotype the close calls, and set fund / fix / kill thresholds in advance.
That is what an AI Bets Audit installs in two weeks, drawing on the same method we have run across 4,000+ enterprise experiments for teams like Tabcorp, AGL, and RACQ. The goal is simple: stop funding the 80-to-95% that were never going to pay off, and back the few that will.
Sources
- More than 80% of AI projects fail, roughly twice the failure rate of non-AI IT projects. RAND Corporation, 2024
- 95% of enterprise generative-AI pilots produce no measurable P&L impact. MIT NANDA, State of AI in Business 2025
- 88% of organisations use AI, but only 39% report any enterprise-level EBIT impact. McKinsey, The State of AI 2025
- Only 22% of companies have moved beyond proof-of-concept, and just 4% create substantial value from AI. BCG, Where’s the Value in AI? 2024
- Over 40% of agentic AI projects will be canceled by the end of 2027, due to unclear business value. Gartner, 2025