Free resource
Most AI projects don't fail because the technology was weak. They stall because the organisation around the technology was never made ready for it. Six causes account for almost all of them — and each one is visible before you spend anything.
You will hear several different numbers in one week, and most of them are real. They differ because they measure different things over different windows. Here is the evidence base as it actually stands.
More than 80% of AI projects fail — roughly twice the failure rate of IT projects that don't involve AI. The figure draws on prior estimates plus 65 structured interviews with experienced AI practitioners.
Summary of the evidenceThe share of companies abandoning most of their AI initiatives rose from 17% to 42% in a year. On average, organisations scrapped 46% of proofs of concept before they reached production.
S&P Global findingsPredicted at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs or unclear business value. Gartner's later analysis puts the realised rate at 50%.
Gartner press release95% of generative AI pilots showed no measurable profit-and-loss impact. The report attributes this to a learning gap in how tools are integrated into workflows, not to model quality.
MIT report coverageDifferent samples, different definitions of "failure" — but the causes behind them are consistent, and they are organisational rather than technical. Those causes are below.
Each cause names the readiness dimension that would have caught it. That is the same set of dimensions the free assessment scores.
The tool is in daily use, but nobody can say who is accountable for its output or for the decision it influenced. When something goes wrong, the search for an owner starts on the day of the incident.
Who signs off this output before it reaches a customer?
Strategy & Leadership · Governance & Risk
The model is fine; the records behind it are not. Duplicates, missing fields and three versions of the same customer mean the output is only ever as trustworthy as the worst input.
If we were asked to show the data behind this output, could we?
Data Readiness
People use AI the way people use anything they were never trained on: unevenly, quietly, and slightly wrongly. The risky habits are the ones nobody was taught to avoid.
Could everyone using AI here describe what must never be pasted into it?
Skills & Culture
There is no decision path: what is allowed, who approves it, what gets logged, and what happens when the output is wrong. Governance exists as a document, not as a route people follow.
What happens today if an AI output is wrong — who finds out, and how?
Governance & Risk
The project started because a competitor did it, or because a demo was impressive. Nobody wrote down what success would look like in numbers, so nobody can call it a failure either.
What number were we trying to move, and who agreed it?
Strategy & Leadership
The pilot worked in a room with four people. Nobody planned what changes when 400 people use it: the process, the exceptions, the training, the ownership, the reporting.
What changes in the team's weekly routine when this goes live?
Processes & Automation
A one-page PDF: the six causes, the question that catches each one, and the readiness dimension behind it. Tick a cause only if you have seen it in your own organisation.
The free assessment scores the six readiness dimensions and shows which of these six causes you are already living with — before a pilot makes it obvious.