MIT report finds 95% of generative AI enterprise pilots show no measurable return

A report from MIT Media Lab’s Project NANDA, “The GenAI Divide: State of AI in Business 2025,” found that despite an estimated $30–40 billion in enterprise spending on generative AI, 95% of surveyed corporate pilots were producing no measurable return on profit and loss. The researchers reviewed more than 300 publicly disclosed AI initiatives, conducted 52 structured interviews with people at companies deploying the technology, and gathered 153 survey responses from senior leaders.

The report’s central claim was a divide rather than uniform failure: about 5% of pilots — generally ones built or customised around a specific workflow rather than generic chatbot deployments — were extracting substantial value, while the rest stalled with no measurable financial impact. It attributed the gap less to model quality than to integration: off-the-shelf tools like ChatGPT worked well for individual employees precisely because of their flexibility, but that same flexibility meant they did not learn from or adapt to a company’s specific processes, and enterprises that treated deployment as a one-off pilot rather than an ongoing integration effort saw returns stall.

The finding was widely reported from mid-August onward and, cited alongside other softening indicators, was linked in press coverage to a brief selloff in AI-related stocks. Critics questioned the report’s methodology, noting companies were often reluctant to disclose failure rates and that a sample built from public disclosures and leader interviews likely skews toward more visible, better-resourced deployments. The 95% figure nonetheless became a widely cited reference point in arguments that enterprise generative AI adoption was running well ahead of demonstrated return.