MIT study reveals 95% of generative AI pilots fail

A new report from MIT's NANDA initiative reveals that while generative AI holds great promise, most enterprise-level initiatives are failing to drive rapid revenue growth. The GenAI Divide: State of AI in Business 2025 research, based on interviews, surveys, and analysis of hundreds of deployments, found that only about 5% of AI pilot programmes make it to full-scale production.

The report's lead author, Aditya Challapally, explained that 95 per cent of larger companies are falling short due to a "learning gap" and flawed integration strategies. Challapally noted that generic tools like ChatGPT, while effective for individuals due to their flexibility, often struggle in enterprise settings because they fail to adapt to complex workflows.

The study also found a major misalignment in spending, with over half of generative AI budgets allocated to sales and marketing despite back-office automation yielding the highest return on investment. The research suggests that companies that purchase specialised AI solutions and form partnerships have a success rate of around 67 per cent, whereas internal builds succeed only a third of the time. The report also highlights the growing use of "shadow AI" and a trend of not backfilling roles in customer support and administrative positions due to automation.

Training Announcement: Freevacy offers a range of independently recognised professional AI governance qualifications and AI Literacy short courses that enable specialist teams to implement robust oversight, benchmark AI governance maturity, and establish a responsible-by-design approach across the entire AI lifecycle. Find out more.