A new report from MIT’s NANDA initiative finds that only about 5% of enterprise generative AI pilot programs achieve rapid revenue growth, with most failing to deliver a measurable impact.

The study analyzed 300 public AI deployments, interviewed 150 business leaders, and surveyed 350 employees.

Report lead author Aditya Challapally said startups that focus on a single use case and partner strategically see more success, while most companies struggle due to poor integration and a “learning gap” around AI tools.

More than half of enterprise generative AI budgets go to sales and ad tools, but the highest returns were found in back-office automation.

The report found that purchasing AI tools from vendors and forming partnerships had a 67% success rate, compared to much lower outcomes for in-house development.

Workforce changes are occurring mostly through not replacing vacated roles, especially in support and administrative positions.

🔗 Source: Fortune

🧠 Food for thought

1️⃣ AI scaling challenges have persisted across different waves of technology

The MIT finding that 95% of generative AI pilots fail to achieve rapid revenue acceleration reflects a consistent pattern in enterprise AI adoption spanning several years.

McKinsey research from 2019 found that only 16% of organizations had successfully scaled AI beyond experimentation, while Accenture reported that 84% of executives believed they couldn’t achieve their business strategy without scaling AI, yet only 16% had moved past the pilot phase23.

Even earlier, Forbes documented that 25% of organizations using AI reported failure rates of up to 50%, primarily due to lack of skilled staff and unrealistic expectations4.

This pattern suggests that the challenges in AI implementation extend beyond the specific technology, whether traditional AI or generative AI, and point to deeper organizational and execution issues that have proven difficult to resolve over time.

2️⃣ External partnerships outperform internal development in AI success rates

The MIT study reveals a striking difference in success rates: purchasing AI tools from specialized vendors succeeds about 67% of the time, while internal builds succeed only one-third as often1.

This finding is particularly relevant as many financial services firms are building proprietary generative AI systems internally, yet the research suggests they may be choosing the path with lower odds of success1.