AI pilots struggle despite hype

A look at why most enterprise pilots stall and how buyers differ from builders.

MIT Report Finds 95 Percent of Generative AI Pilots Fail

A new MIT GenAI Divide 2025 report shows that only about 5 percent of enterprise generative AI pilots deliver rapid revenue gains. The study is based on 150 interviews, a survey of 350 employees, and analysis of 300 public deployments. It finds a sharp gap between companies that buy AI tools from vendors and those that build in house, with the main hurdle being how tools fit into real work. The research points to a learning gap for both tools and organizations, rather than flaws in the models themselves. Budgets also tilt toward sales and marketing tools while the biggest returns come from back office automation and process improvement.

The report then outlines what drives success. Empowering line managers to drive adoption, ensuring deep integration with existing systems, and choosing tools that adapt over time are key. Purchases and partnerships outperform internal builds by about two to one, especially in regulated sectors like finance where many firms still pursue bespoke systems. Shadow AI and the difficulty of measuring true productivity remain persistent challenges as organizations scale AI use. Looking ahead, early pilots are testing agentic AI that can learn and act within set boundaries, hinting at the next phase of enterprise deployment.

The findings force a reckoning with how firms define success for AI. Too many budgets chase flashy front end tools while the real value lies in streamlining back office work and enforcing governance. The lesson is not to abandon AI but to redesign adoption as a program that couples technology with people and processes. Firms that succeed tend to decentralize decision making to line managers and invest in tools that genuinely integrate with existing workflows.

Governance and talent emerge as the unseen bottlenecks. Shadow AI creates risk when sanctioned controls lag behind usage, and leadership must close the gap between experimentation and accountable operations. As agents grow more capable, firms will need clear boundaries and continuous learning to avoid missteps. The MIT study suggests a pragmatic path: buy wisely, partner where possible, and measure real impact beyond vanity metrics.

Highlights

The next phase of enterprise AI will reward disciplined adoption over hype.