AI concepts

GenAI Divide (the gap in generative AI deployments)

A concept introduced by MIT Project NANDA in a July 2025 report describing the split between the five percent of firms that achieve a measurable return from generative AI deployments and the ninety-five percent whose pilots end with no impact on the P&L. After spending thirty to forty billion dollars.

Primary source: MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, lipiec 2025

MIT NANDA published the report after analysing 300 GenAI deployments in US corporations and 150 structured interviews with transformation leaders. The 95 percent figure became the most frequently cited AI statistic of 2025. It was also the most frequently misread.

What the study actually measures

Success was defined strictly: a lasting, measurable impact on the P&L, confirmed by end users and the management team, sustained for at least six months in production. Pilots that ended with a rollout to a single department do not qualify. Pilots with positive business signals but without consensus between operations and the board do not qualify.

Methodological critique

Sean Goedecke, an independent analyst of enterprise technology, in a September 2025 analysis, notes that 60 percent of the firms in the MIT sample even considered a GenAI deployment, 20 percent built a pilot, and 5 percent achieved measurable impact. That gives a success rate of 8.3 percent of attempts, not 5 percent. The 95 percent figure applies only to firms that moved into the pilot phase.

A second point of critique: 84 percent of IT transformations fail in general, according to the CHAOS research by the Standish Group from 2024. IT projects above 10 million dollars have a 98 percent non-completion rate. GenAI is not uniquely hard, it is uniquely visible.

What sets the 5 to 20 percent of successes apart

Four repeatable patterns, identical across five independent studies from 2024 and 2025 (RAND, McKinsey, BCG, MIT, Gartner):

  • workflow-first design, processes designed before the model is chosen
  • data integration priority, integrating data before the pilot
  • realistic scoping, one problem instead of a strategic transformation
  • sustained C-level involvement, sponsorship through the whole cycle, not just at the start

We diagnose which group your company is in within the AI Readiness Audit. A full analysis of the phenomenon is in the pillar article.