A widely repeated statistic about AI project failure, cited across dozens of business publications and attributed to the RAND Corporation, does not appear anywhere in RAND's actual report. It was invented in secondary coverage and passed along uncritically until it became one of the most-quoted numbers in enterprise AI journalism.

63%
Organizations reporting no enterprise-level EBIT impact from AI, per McKinsey's 2026 survey of 1,993 respondents
89%
Executives reporting no measurable productivity impact from AI, per a rigorous NBER survey of nearly 6,000 firms
54%
C-suite executives who say AI adoption is tearing their company apart, per WRITER's 2026 survey
20%
The cleanest measured project-failure rate available, per Gartner's 2026 survey of 782 I&O leaders

Two Most Popular AI Failure Rate Claims Misrepresent the Findings

The most-cited AI failure statistic in circulation, "more than 80% of AI projects fail," traces back to a 2024 RAND Corporation study. RAND's underlying research was 65 qualitative interviews with data scientists and engineers about the root causes of AI project failure. There is no denominator of projects in it, no measured percentage of RAND's own. The 80% figure is RAND citing prior literature, not a finding RAND produced. A further breakdown that has circulated widely, that 33.8% of projects are abandoned before production, 28.4% reach production but fail to deliver value, and 18.1% run but never recoup costs, does not appear in RAND's report at all. It was fabricated in secondary coverage and repeated as if it were sourced.

MIT Project NANDAThe other figure that dominates this conversation, that 95% of organizations deploying generative AI see zero measurable return, is accurate but far narrower than how it's usually cited. It measures generative AI pilots with no measurable P&L impact within a six-month window, drawn from a directional, conference-attendee sample. Treated as shorthand for "AI doesn't work," it overstates what the underlying research found. Treated correctly, as one data point about short-horizon financial visibility into early pilots, it's a legitimate and useful figure.

A third problem sits underneath both of these: Even the legitimate surveys measure fundamentally different things and get cited as if they're interchangeable. Project cancellation, missed EBIT targets, stalled pilots, and individual workload errors all get called "AI failure" by different studies, with failure rates ranging from 20% to 95% depending entirely on which of those categories a given survey is measuring. Citing any of them without naming which category they belong to is how a defensible statistic turns into a misleading headline.

Correct the Numbers, and the Picture Doesn't Change

None of that misrepresentation makes the aggregate picture wrong. Set the fabricated RAND breakdown and MIT's overstretched 95% aside entirely, and rely only on the cleanest, most methodologically careful figures available, and the story doesn't improve.

The single most rigorous data point in this entire debate comes from outside the survey-research industry altogether: A National Bureau of Economic Research working paper, co-authored by Stanford economist Nicholas Bloom and fielded with central bank researchers at the Federal Reserve Bank of Atlanta, the Bank of England, and the Deutsche Bundesbank, surveyed nearly 6,000 senior executives across the US, UK, Germany, and Australia and found that 89% report no impact on labor productivity from AI over the past three years, and more than 90% report no impact on employment. This is not a marketing-adjacent survey with an interest in the answer either way. It's academic labor economics, and it lands on the same conclusion as the industry surveys below.

Gartner's April 2026 survey of 782 I&O leaders found 20% of AI use cases fail outright, the single cleanest project-failure figure available among the industry surveys, and separately found only 28% of AI infrastructure use cases fully succeed and meet ROI expectations. McKinsey's 2026 survey of 1,993 respondents, one of the largest and most rigorous in this space, found 63% of organizations report no enterprise-level EBIT impact from AI at all. Domino Data Lab's April 2026 survey of 639 AI leaders found 57% report ROI failing to outpace investment over a two-year window, a hard financial definition, not a subjective one.

None of these figures depend on RAND's fabricated breakdown or MIT's narrow six-month window. They arrive at the same basic conclusion independently: A minority of AI deployments are delivering clear, measurable financial return, and a majority are not.

Productivity Improvement Findings Shrink Under Scrutiny

Surveys showing improved ROI and productivity, like the RAND and MIT results, are less impressive when scrutinized. KPMG's Q2 2026 survey found 76% of leaders say AI is delivering meaningful business value, up 12 points from Q1's 64%. Dun & Bradstreet found 60% of businesses report at least some measurable ROI in 2026, a sharp rise from single digits in 2025. Domino Data Lab found 93% report improved production capability, up from 88% the year before.

But look closely at what the improving numbers say. Dun & Bradstreet's 60% reporting "at least some measurable ROI" includes many organizations with marginal, barely-there returns; only 24% report broad or strong returns specifically. KPMG's rising value-perception figure measures sentiment, whether leaders believe AI is delivering value, not a verified financial outcome. Improving sentiment and improving capability are positive trends. They are not the same claim as "most AI deployments now deliver strong ROI," and none of the 2026 data supports that stronger claim. The trend line holds up. The majority of deployments still landing short of clear, strong returns holds up too, and the two facts sit alongside each other rather than cancelling out.

One Piece of Where the Productivity Goes

AI fails to produce measurable ROI for many distinct reasons, and no single study explains all of them. Agentic AI projects fail for their own reasons, unreliable tool use, weak guardrails, tasks too complex to hand off safely. Other automation initiatives fail for reasons that have nothing to do with individual time savings at all: bad data, unclear ownership, integration costs that exceed the budget, or a use case that was never well matched to the technology in the first place. The NBER finding doesn't specify which of these is responsible for its 89% figure.

What a separate study, Boston Consulting Group's fourth annual Global AI at Work survey of nearly 12,000 employees, offers is not an explanation for all of it. It's one clue, worth taking seriously precisely because it's measurable. It found that 42% of regular AI users are personally saving a full workday every week. That figure doesn't resolve NBER's 89%. It's one piece of the picture: At least some of the productivity that isn't showing up at the company level is showing up at the individual level instead, unconverted.

BCG, Global AI at Work 2026Of the employees saving that time, 66% report getting little or no guidance on what to do with it, and more than half say they aren't redirecting the saved hours into anything their organization would recognize as strategic work. BCG's reporting on the survey states that an explicit strategy for redirecting that time lifts AI's measured business impact by 25 percentage points, compared with roughly 5 points from giving people better tools alone. Where AI's ROI gap is caused by unconverted time rather than a failed deployment, the bottleneck isn't the technology itself.

Read together, NBER and BCG describe part of the same gap from two different altitudes, not the whole of it. NBER measured whether AI shows up in company-wide productivity and employment statistics, and mostly, it doesn't yet. BCG measured one specific reason that might be true: substantial individual time savings that are scattering into busywork rather than concentrating into anything a finance department would register as output.

What the WRITER Survey Adds That the ROI Numbers Alone Don't

Every figure above measures outcomes. WRITER's 2026 Enterprise AI Adoption survey, run with independent research firm Workplace Intelligence across 1,200 non-technical employees and 1,200 C-suite executives, measures something closer to lived experience inside the organizations running these deployments, and it's bleaker than the ROI statistics alone convey.

WRITER, 2026 Enterprise AI Adoption Survey97% of executives say their company deployed AI agents in the past year. 79% of organizations report challenges adopting AI, a double-digit increase over 2025. And 54% of C-suite executives admit, in their own words, that AI adoption "is tearing their company apart." Nearly universal deployment and a majority of the people responsible for that deployment describing significant organizational strain are the same finding, stated twice.

WRITER's findings reinforce the endemic AI failure rate among organizations from different angles and from the vantage point of the C-suite. On ROI specifically: Few leaders say they've seen significant ROI from generative AI (29%) or AI agents (23%), and nearly half (48%) feel that AI adoption at their company has been a massive disappointment.

On strategy: 39% of C-suite leaders admit they don't have a formal strategy in place to drive revenue from AI tools. Even where strategies do exist, quality is lacking, 75% of executives say their company's AI strategy is more for show than for actual internal guidance.

On internal friction: 78% of executives say AI has created tension between IT and other lines of business, with 55% reporting that AI use is a chaotic free-for-all at their company.

And in contrast to the employee time savings the BCG survey found, WRITER finds active resistance rather than quiet inefficiency. Instead of embracing AI, some workers are pushing back: 29% of employees, including 44% of Gen Z, admit to sabotaging their company's AI strategy, for example by entering company information into public tools, using unapproved tools, or refusing to use AI. Executives recognize the danger: 76% say employee sabotage poses a serious threat to their company's future.

That detail matters because it closes a gap the pure ROI numbers leave open. A survey measuring EBIT impact can be technically accurate while still missing what's happening on the ground, the politics, the exhaustion, the sense among leadership that the technology is moving faster than the organization can absorb it. WRITER's finding doesn't contradict McKinsey's or Domino's. It explains what the shortfall feels like from inside the C-suite making the deployment decisions.

The bottom line: The AI failure narrative has a documented misrepresentation problem, a fabricated RAND statistic and an overstretched MIT figure have both been repeated far past what the underlying research supports, and that's worth correcting every time it comes up. But correcting the citations doesn't rescue the underlying story. The cleanest, most carefully measured figures available, independent of the discredited ones, still show a minority of AI deployments delivering clear financial return, a conclusion an NBER labor-economics survey of nearly 6,000 executives reaches just as firmly as the industry research does.

BCG's employee-level data offers one partial explanation: Substantial time savings exist for a meaningful share of employees, but a lot of that time is leaking out of organizations that never decided what to do with it. It's a contributing factor, not the whole story, agentic AI projects and other automation efforts fail for their own separate reasons. The improving sentiment numbers hold up but describe a shift within that minority, not a reversal of it.

And the WRITER survey's finding that most C-suite executives describe significant organizational strain from AI adoption is the clearest signal that the ROI gap isn't a statistics problem. It's a deployment problem, and it's still the majority experience.

Sources: AI Pulse · Where This Breaks · workplaceai.ai. The RAND Corporation study and the identification of the fabricated breakdown attributed to it: RAND Corporation report RRA2680-1, "The Root Causes of Failure for AI Projects," with the fabrication identified via governanceai.io. MIT Project NANDA's GenAI Divide report and its six-month P&L-impact definition: MIT NANDA, "The GenAI Divide: State of AI in Business 2025." NBER Working Paper 34836, "Firm Data on AI," by Ivan Yotzov, Jose Maria Barrero, Nicholas Bloom, Philip Bunn, Steven J. Davis, and co-authors, fielded November 2025 through January 2026. Gartner's I&O failure figures: Gartner's April 2026 survey of 782 I&O leaders. McKinsey's EBIT-impact figures: McKinsey's 2025 and 2026 State of AI surveys, 1,993 respondents. Domino Data Lab's ROI figures: Domino Data Lab, April 2026 survey of 639 AI leaders. The counter-narrative figures: KPMG's Q2 2026 survey of 2,145 senior leaders; Dun & Bradstreet's 2026 survey of 10,000 businesses; Domino Data Lab, same 2026 survey. BCG's employee-level findings, including the 25-point and 5-point strategy-versus-tools figures: Boston Consulting Group's fourth annual Global AI at Work survey, published via BCG's press release, "AI Is Reshaping Jobs Faster Than Companies Are Reshaping Work" (June 3, 2026), and the accompanying report, "AI at Work: Why Strategy Matters More Than Tools," nearly 12,000 respondents. The WRITER survey: WRITER's 2026 Enterprise AI Adoption survey, conducted with Workplace Intelligence, 1,200 employees and 1,200 C-suite executives. Every figure above is attributed to its original reporting; none is a WorkplaceAI study.