MIT's claim that 95% of AI pilots fail - Is this true?

MIT’s August 2025 preliminary NANDA report claimed that 95% of enterprise generative-AI pilots failed to deliver measurable profit-and-loss impact. A February 2026 critique by 80,000 Hours’ Rob Wiblin said the figure treated organizations that had not run custom, task-specific pilots as failures; the report drew on about 300 disclosed initiatives, 52 interviews and 153 conference survey responses. The report’s authors also did not disclose that several developed and sold agentic-AI infrastructure, according to the critique. Later surveys cited by the source indicate widespread AI use but limited measured financial returns. McKinsey’s May-June 2026 survey of 1,719 participants in 97 countries found 88% of organizations used AI regularly in at least one function, while 37% reported a positive EBIT contribution and 6% attributed at least 5% of operating profit to AI. Eighty percent of respondents said AI made their work faster. McKinsey found high-performing organizations redesigned workflows around AI about three times as often as others. Deloitte found 25% of organizations had moved at least 40% of AI pilots into production, while Capgemini reported 66% saw measurable productivity or decision-quality gains. UK ONS data cited in the source found 35% of businesses used at least one AI technology, with only 10% describing use as extensive.