MIT study reveals why most enterprise AI pilots fail

MIT's research confirms what many of us suspect—95% of enterprise generative AI pilots are failing to drive meaningful revenue impact. The study tracked 150 leadership interviews and 300 public deployments, so this isn't anecdotal handwaving.

The findings are more than model performance issues. Companies face what MIT calls the "learning gap"—generic tools like ChatGPT work brilliantly for individuals because they adapt to personal workflow, but they can't learn from organizational patterns or integrate into enterprise processes. So they stall.

Most companies are also betting on the wrong applications. Over half of AI budgets flow to sales and marketing tools, but MIT found the strongest ROI comes from back-office automation—eliminating outsourcing costs and streamlining operations where workflow integration matters most.

But there is a different path, and based on my experience with partners over my career, I’m not surprised. Companies that buy specialized tools and build vendor partnerships succeed 67% of the time, while internal builds succeed only one-third as often. This challenges the reflex in highly regulated sectors to build everything in-house. Line managers need to drive adoption rather than centralizing everything in AI labs, and tools need deep integration capabilities rather than broad generalist features.

For teams navigating AI strategy, this data points toward partnership over proprietary development and workflow integration over feature breadth. The winners aren't building the smartest tools—they're choosing the ones that learn from how work gets done.

https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/