MIT Report: 95% of Company AI Pilots Are Failing

A report by MIT’s NANDA initiative, titled “The GenAI Divide: State of AI in Business 2025,” reveals that despite the potential generative AI holds for enterprises, 95% of corporate AI pilot projects fail to deliver financial results. The research unveils the reasons behind the success stories and stagnant projects.

Many companies assume AI projects fail due to performance or regulation of the model, but MIT’s research suggests the main problem lies in the “learning gap.”

Another issue is resource misallocation, where over half of AI budgets go to sales and marketing tools. However, the report shows that AI tools for back-office automation, which can help eliminate business process outsourcing, cut external agency costs, and streamline operations, offer the biggest return on investment (ROI).

Analysis of the 5% of projects where implementing AI pilot tools has resulted in rapid revenue acceleration has shown that success lies in the AI adoption method of choice. Using tools from specialized external vendors has proved successful 67% of the time, while internal AI builds have been successful only one-third as often. This is especially true for companies in finance and other highly regulated sectors. Due to safety concerns, many organizations in such industries opt to develop proprietary generative AI solutions.

Aditya Challapally, the lead author of the report, told Fortune that while generic tools like ChatGPT excel for individual use, they often fail in enterprise settings. This is due to their inability to adapt to complex, company-specific workflows, unless they undergo specific training. That’s why it’s important to invest in enterprise-grade tools that integrate seamlessly and can be adapted to evolving business needs over time.

The phenomenon of employees using unsanctioned tools like ChatGPT at an increased rate is known as “shadow AI.” The term refers to using AI tools for work without company oversight, which not only poses security risks but also makes it impossible to measure their impact on productivity and profit.

The report also suggests another key driver of successful AI adoption is giving line managers – not only central AI teams – the authority to lead implementation.

Despite failing to deliver financial gains, most AI pilot initiatives are disrupting the workforce landscape. The report suggests that AI adoption is affecting typically outsourced roles, like customer support and administrative assistance, the most.