A new report from MIT’s NANDA initiative, published today, pours cold water on the narrative that enterprise generative AI is driving immediate returns. The study — an analysis of 300 public AI deployments, based on 150 interviews with leaders and a survey of 350 employees — finds that 95% of enterprise AI pilots deliver zero measurable impact on the P&L.

The failure, according to lead author Aditya Challapally, is not a model-quality problem but an organization one. Generic tools like ChatGPT excel for individuals because of their flexibility, but they stall in enterprise use since they don’t learn from or adapt to workflows — a phenomenon the report calls a “learning gap.” Only about 5% of organizations are getting a pilot across to real, measurable operational or financial impact, and the ones that do tend to buy from specialized vendors rather than build in-house. “Almost everywhere we went, enterprises were trying to build their own tool,” Challapally said, but the data showed building succeeds only one-third as often as buying.

The report also reveals a misalignment in resource allocation: more than half of generative AI budgets go to sales and marketing tools, while the biggest ROI comes from back-office automation, such as eliminating business process outsourcing and cutting external agency costs. Workforce disruption is already underway, mostly through attrition rather than layoffs, concentrated in customer support and administrative roles.

The report also highlights the widespread use of ‘shadow AI’ — unsanctioned tools like ChatGPT — and the ongoing challenge of measuring AI’s impact on productivity and profit.