AI Project Failure Rate Statistics (2026): What Eight Studies Measure

Published AI failure figures range from 30% to 95%. They differ because each study counts a different thing: pilots without profit impact, proofs of concept that never ship, abandoned programs or forecast cancellations.

Published October 2026

Key findings

MIT NANDA found 95% of companies in its data set getting little or no measurable profit-and-loss impact from generative AI pilots.
IDC, working with Lenovo, found that for every 33 AI proofs of concept a company launched, only four reached production, an 88% drop-off.
S&P Global found 42% of companies abandoning most of their AI initiatives in its 2025 survey, up from 17% a year earlier.
In RAND interviews, 30 of 50 industry data scientists and engineers named persistent data quality problems as a cause of failure.
Gartner forecasts that over 40% of agentic AI projects will be canceled by the end of 2027.

“Most AI projects fail” is one of the most repeated lines in enterprise technology. The figure attached to it changes depending on who is quoted. We collected the eight most cited studies and recorded, for each, what was measured, who was asked and when. Read side by side, they describe several different stages where AI work stops.

Failure figures by study

Horizontal bar chart of eight published AI failure figures from 30% to 95%, labeled by study and what each measures — Sources listed below. Each bar measures a different outcome.

Published AI failure statistics and what they measure
Study	Figure	What it measures	Basis
MIT NANDA, The GenAI Divide (2025)	95%	Companies whose generative AI pilots show little or no measurable P&L impact; about 5% of pilots reach rapid revenue acceleration	150 leader interviews, a survey of 350 employees and 300 public deployments
IDC with Lenovo (2025)	88%	Proofs of concept that did not reach wide-scale deployment (4 of every 33)	IDC research reported by CIO.com
RAND, Ryseff, De Bruhl and Newberry (2024)	More than 80%	AI projects estimated to fail, about twice the rate for IT projects without AI; cited from prior estimates	Literature, with 65 interviews (50 industry, 15 academic)
BCG, Where’s the Value in AI? (2024)	74%	Companies that have yet to show tangible value from AI	Survey of 1,000 executives in 59 countries
S&P Global Market Intelligence (2025)	46%	Average share of AI proofs of concept scrapped before production	Survey of 1,006 IT and business professionals in North America and Europe
S&P Global Market Intelligence (2025)	42%	Companies abandoning most of their AI initiatives (17% in 2024)	Same survey
Gartner (June 2025)	Over 40%	Agentic AI projects forecast to be canceled by end of 2027	Analyst forecast
Gartner (July 2024)	At least 30%	Generative AI projects forecast to be abandoned after proof of concept by end of 2025	Analyst forecast

The high figures measure value. MIT’s 95% and BCG’s 74% count companies or pilots that have not produced measurable financial returns, which includes projects still running. The middle figures measure the step from proof of concept to production: IDC’s 88% and S&P Global’s 46% average. The Gartner figures are forecasts of cancellation, not observed outcomes. RAND’s “more than 80%” is an estimate the authors cite from earlier sources, not a count from their interviews.

Abandonment rose between 2024 and 2025

Bar chart showing the share of companies abandoning most AI initiatives rising from 17% in 2024 to 42% in 2025 — Source: S&P Global Market Intelligence, Voice of the Enterprise: AI & Machine Learning, Use Cases 2025.

S&P Global’s survey is the only source here that measures the same question twice. The share of companies abandoning most of their AI initiatives rose from 17% to 42% in one year, and respondents named cost, data privacy and security risks as the top obstacles, according to CFO Dive’s summary. Gartner gave similar reasons for its forecasts: poor data quality, inadequate risk controls, escalating costs or unclear business value in 2024, and escalating costs, unclear business value or inadequate risk controls for agentic projects in 2025.

The causes practitioners name

Bar chart of failure causes named by RAND's 50 industry interviewees, led by data quality at 30 of 50 — Source: RAND RR-A2680-1, August 2024.

Causes raised by RAND’s industry interviewees
Issue raised	Interviewees (of 50)
Persistent data quality issues	30
Failures driven by data scientists pursuing new tools and techniques	16
Senior leaders underestimating the time needed to train a model	14
Lack of domain understanding on the technical team	10
Rigid interpretations of agile development	10

RAND interviewed 50 data scientists and engineers in industry and 15 academics. Data quality came up most often. BCG reached a similar conclusion from the executive side: about 70% of the challenges companies reported in AI programs came from people and process issues, 20% from technology and 10% from the algorithms themselves.

What the same studies say succeeds

MIT NANDA reported that buying AI tools from specialized vendors and building partnerships succeeded about 67% of the time, while internal builds succeeded about one-third as often. BCG found that the 26% of companies it classed as AI leaders pursued about half as many opportunities as their peers and put 70% of their resources into people and processes. IDC’s 4-in-33 production rate is the more useful planning figure for a team deciding how many pilots to fund.

Sources (8)