{
  "post": {
    "title": "\"95% of AI pilots fail\": what the MIT report actually measured",
    "url": "https://bardak.ai/posts/mit-95-percent-ai-pilots-fail",
    "updated": "2026-10-29T15:31:36.978Z"
  },
  "citation": "\"95% of AI pilots fail\": what the MIT report actually measured, Ivan Bardak, bardak.ai, https://bardak.ai/posts/mit-95-percent-ai-pilots-fail",
  "datasets": [
    {
      "key": "what-the-mit-nanda-report-is-built-on",
      "title": "What the MIT NANDA report is built on",
      "type": "benchmarkTable",
      "source": "Challapally, Pease, Raskar and Chari, The GenAI Divide: State of AI in Business 2025, pages 2, 6, 24 and 25",
      "caption": "The report mentions bootstrapped confidence intervals but shows none, and gives no response rates or breakdown of the sample by industry or size.",
      "csv": "https://bardak.ai/data/mit-95-percent-ai-pilots-fail/what-the-mit-nanda-report-is-built-on.csv",
      "header": [
        "Source of evidence",
        "Size",
        "What to know"
      ],
      "rows": [
        [
          "Review of publicly disclosed AI initiatives",
          "Over 300",
          "Public announcements, which skew towards what companies choose to publicise"
        ],
        [
          "Structured interviews",
          "52 organisations",
          "The basis for the main figures; the report calls them directional"
        ],
        [
          "Survey",
          "153 senior leaders at four industry conferences",
          "Called a lightweight supplement; conferences not named"
        ],
        [
          "Time window",
          "January to June 2025; ROI judged six months after the pilot",
          "Six months may be too short for complex systems, as the report itself says"
        ]
      ]
    },
    {
      "key": "what-the-95-refers-to-page-by-page",
      "title": "What the 95% refers to, page by page",
      "type": "benchmarkTable",
      "source": "MIT NANDA (2025), pages 3, 6 and 7",
      "caption": "Organisations, pilots, tools and solutions are not the same population. A company with one failed pilot and one success counts differently in each.",
      "csv": "https://bardak.ai/data/mit-95-percent-ai-pilots-fail/what-the-95-refers-to-page-by-page.csv",
      "header": [
        "Page",
        "The claim",
        "Unit"
      ],
      "rows": [
        [
          "3",
          "95% of organisations are getting zero return on generative AI",
          "Organisations"
        ],
        [
          "3",
          "Only 5% of integrated AI pilots are extracting millions in value",
          "Pilots"
        ],
        [
          "6",
          "Only 5% of custom enterprise AI tools reach production",
          "Tools"
        ],
        [
          "7",
          "A 95% failure rate for enterprise AI solutions",
          "Solutions"
        ]
      ]
    },
    {
      "key": "mit-nanda-share-of-organisations-at-each-stage",
      "title": "MIT NANDA: share of organisations at each stage",
      "type": "chart",
      "source": "MIT NANDA, The GenAI Divide: State of AI in Business 2025, page 6",
      "caption": "General-purpose tools are chat assistants such as ChatGPT and Copilot; task-specific systems are built into a workflow, custom or bought. The authors call the figures directionally accurate, based on interviews rather than company reporting.",
      "csv": "https://bardak.ai/data/mit-95-percent-ai-pilots-fail/mit-nanda-share-of-organisations-at-each-stage.csv",
      "header": [
        "stage",
        "General-purpose tools",
        "Task-specific systems"
      ],
      "rows": [
        [
          "Investigated",
          "80",
          "60"
        ],
        [
          "Piloted",
          "50",
          "20"
        ],
        [
          "Successfully implemented",
          "40",
          "5"
        ]
      ]
    },
    {
      "key": "other-surveys-on-value-from-ai",
      "title": "Other surveys on value from AI",
      "type": "benchmarkTable",
      "source": "MIT NANDA 2025; Gartner 2026; IBM 2025; BCG 2024; McKinsey 2026",
      "caption": "None of these is a random sample of companies and every figure is self-reported. They agree that clear value is the exception so far; they do not support a 95% failure rate.",
      "csv": "https://bardak.ai/data/mit-95-percent-ai-pilots-fail/other-surveys-on-value-from-ai.csv",
      "header": [
        "Survey",
        "Sample",
        "Finding",
        "What it measures"
      ],
      "rows": [
        [
          "MIT NANDA, July 2025",
          "52 interviews, 153 survey responses",
          "5% of task-specific systems reached production with sustained impact",
          "Self-reported impact, interview sample"
        ],
        [
          "Gartner CIO survey, October 2026",
          "2,002 CIOs, May to July 2026",
          "13% say their enterprise has realised significant or transformational value from AI",
          "CIO judgement of value"
        ],
        [
          "IBM CEO Study, May 2025",
          "2,000 CEOs, February to April 2025",
          "25% of AI initiatives delivered the expected ROI; 16% scaled enterprise-wide",
          "CEO estimates, all AI"
        ],
        [
          "BCG, October 2024",
          "1,000 executives in 59 countries",
          "26% have the capabilities to move beyond proofs of concept and generate tangible value",
          "Self-assessed capability"
        ],
        [
          "McKinsey, August 2026",
          "1,719 respondents in 97 countries",
          "37% report any EBIT contribution from AI; about 6% see 5% of EBIT or more",
          "Self-reported financial impact, all AI"
        ]
      ]
    },
    {
      "key": "appraising-the-evidence",
      "title": "Appraising the evidence",
      "type": "benchmarkTable",
      "source": "",
      "caption": "",
      "csv": "https://bardak.ai/data/mit-95-percent-ai-pilots-fail/appraising-the-evidence.csv",
      "header": [
        "Evidence",
        "Strength",
        "Main limitation"
      ],
      "rows": [
        [
          "MIT NANDA report",
          "Useful qualitative observations",
          "Preliminary, not peer-reviewed, small interview sample, inconsistent definitions"
        ],
        [
          "Large executive surveys (McKinsey, IBM, BCG, Gartner)",
          "Large samples, repeated over years",
          "Self-reported, non-random, different definitions of value"
        ],
        [
          "My client work",
          "Direct experience of where pilots stall",
          "One practitioner"
        ]
      ]
    }
  ]
}