Source: The GenAI Divide: State of AI in Business 2025 (MIT NANDA)
MIT NANDA's 2025 preliminary report finding that about 5% of enterprise GenAI pilots produce measurable P&L impact, blaming a 'learning gap' in integration rather than model quality.
Contents
- What: preliminary research report (read via Fortune’s coverage; the report itself is gated)
- Publisher: MIT NANDA initiative; lead author Aditya Challapally
- Date: July 2025 report; coverage 2025-08-18
- Original: Fortune: MIT report, 95% of generative AI pilots failing
Summary
Surveys enterprise GenAI adoption and finds a sharp divide: a small minority of pilots produce rapid revenue or cost impact while most show none. Attributes the gap to tools that do not integrate with or learn from company workflows.
Key claims
- Method (as reported by Fortune): about 150 leader interviews, a survey of 350 employees and analysis of 300 public AI deployments. The report’s own methods note (January-June 2025) describes a review of 300+ public AI initiatives, structured interviews with representatives of 52 organisations and 153 senior-leader survey responses from industry conferences; the two descriptions do not match exactly.
- About 5% of pilots produce rapid revenue acceleration; the rest show little or no measurable P&L effect.
- Buying from or partnering with specialised vendors succeeds about 67% of the time; internal builds succeed about one third as often.
- The core barrier is a “learning gap”: generic tools help individuals but stall in enterprise workflows because they do not retain feedback or adapt to context.
- Over half of GenAI budgets go to sales and marketing, but the largest returns come from back-office automation (cutting BPO, agency spend, operations).
- Empowering line managers, not only central AI labs, drives adoption; startups focused on one pain point see strong results.
- “Shadow AI” (unsanctioned personal ChatGPT use) is widespread.
- Caveat: small, anonymised sample in a preliminary report; the 95% headline is often quoted with more certainty than the data supports.