MIT NANDA

MIT NANDA's study "The GenAI Divide: State of AI in Business 2025" reports that 95% of enterprise AI pilots deliver no measurable P&L impact, despite $30-40 billion invested. The methodology behind that number: 52 executive interviews, 153 surveys, and analysis of 300 public deployments. Only 5% of pilots create significant value. The funnel for enterprise-grade systems is narrower still: 60% evaluate, 20% pilot, 5% reach production.

The reason the figure keeps resurfacing is the diagnosis attached to it. MIT NANDA places the bottleneck in organization rather than technology, which moves the lever from IT to HR. The report's own prescription follows from that: shift from building to buying, from central labs to empowered teams, from static tools to adaptive systems. Externally built vendor tools succeed twice as often as internal builds.

A second finding travels with the first. MIT NANDA documents a "shadow AI economy": 40% of companies hold official LLM subscriptions while 90% of employees use personal tools daily, often faster and better than the corporate alternative.

The 95% number now functions as a fixed point other work argues against. Google Cloud's DORA report on the ROI of AI-assisted software development, published April 2026, cites it explicitly while reaching a different conclusion, modelling a 39% ROI and an eight-month payback on a 500-FTE sample and framing AI as an amplifier of whatever organizational conditions already exist.