Is it true that 95 percent of AI pilots fail?
What the MIT NANDA study measured, why the 95 percent figure travelled, what larger surveys found, and the lesson it holds for anyone funding AI.
- Topic
- AI ROI
- Product
- Novie Control
- Format
- Guide
- Reading time
- 2 min
- Published
- 5 November 2026
The 95 percent figure comes from MIT NANDA's 2025 report The GenAI Divide. It found that 95 percent of the organisations in its sample saw no measurable profit and loss impact from generative AI pilots within six months. The sample was 52 interviews and 153 survey responses, and the authors called the findings preliminary. It describes early pilots under a strict test, and larger surveys find many more organisations reporting value.
What the study measured.
Success meant a measurable effect on profit and loss within six months of a pilot. That is a high bar. Deloitte found that most organisations expect a typical AI use case to take two to four years to pay back. A pilot that saves hours, improves quality or is still being scaled at month six counts as a failure under the MIT definition.
TakeawayThe study measured profit and loss impact within six months, which most AI work is not designed to show.
Why the figure travelled.
Fortune reported the study in August 2025, and the number spread quickly, often shortened to "95 percent of AI projects fail". Fortune's follow-up noted the small sample, the six-month window and that NANDA itself works on agent infrastructure. None of that changes what the study did find, which is useful.
TakeawayThe shortened version drops the sample, the definition and the time window.
What larger surveys found.
Wharton and GBK Collective surveyed more than 800 US enterprise decision-makers in 2025: 74 percent of those that measure generative AI ROI reported a positive return. McKinsey found 39 percent attribute some EBIT impact to AI. BCG's 2026 index found almost half of companies now generate value. The surveys ask different questions, and none of them contradicts the core of the MIT finding: a quick, measurable financial return is rare.
TakeawayLarger surveys find more organisations reporting value, and agree that fast financial returns are rare.
The lesson worth keeping.
The MIT report found that back-office automation showed stronger returns than the sales and marketing tools that took about half of budgets, and that pilots stalled when tools did not fit the workflow. The practical point for anyone funding AI is to agree before a pilot starts what return it must show, by when, and what happens if it does not.
TakeawayAgree the return a pilot must show, and by when, before it starts.
How Novie Control does it
You can see what AI costs to build.Can you see what it costs to run?
Sources
- Fortune, MIT report: 95% of generative AI pilots at companies are failing (18 August 2025)
- Fortune, An MIT report that 95% of AI pilots fail spooked investors (21 August 2025)
- Wharton and GBK Collective, Accountable Acceleration: Gen AI Fast-Tracks Into the Enterprise (October 2025)
- McKinsey, The state of AI in 2025: Agents, innovation, and transformation (November 2025)
- BCG, AI Is Starting to Pay Off. Almost 50% of Companies Now Generate Value with It (30 September 2026)
- Deloitte, AI ROI: The paradox of rising investment and elusive returns (22 October 2025)