MIT report: 95% of generative AI pilots at companies are failing

A new study, The GenAI Divide: State of AI in Business 2025, published by MIT’s NANDA initiative, shows that while generative AI carries strong potential for enterprises, the majority of projects are struggling to deliver tangible results.

Despite the global race to integrate advanced AI systems, only about 5% of pilot programs have achieved rapid revenue growth. The rest have either plateaued or shown little to no measurable impact on financial outcomes. The findings are based on 150 leadership interviews, a survey of 350 employees, and an analysis of 300 public AI deployments—painting a clear divide between high-profile success stories and stalled initiatives.

Startups Find Success While Enterprises Struggle

Aditya Challapally, lead author of the report and research contributor at MIT’s project NANDA, emphasized that agility is the difference-maker.

“Some large companies’ pilots and younger startups are really excelling with generative AI,” he noted. For instance, startups led by 19- and 20-year-old founders have scaled revenues from zero to $20 million in just one year. Their success, Challapally explained, comes from focusing on a single pain point, executing precisely, and partnering with enterprises that adopt their solutions effectively.

In contrast, 95% of companies in the study failed to deliver expected outcomes. The main barrier was not the sophistication of the models but the “learning gap”—both for the tools and the organizations using them. While executives frequently cited regulations or model quality as obstacles, MIT researchers found that flawed enterprise integration was the core issue.

Generic platforms like ChatGPT work well for individuals due to their adaptability. But within corporations, they falter since they don’t seamlessly learn from or integrate into business workflows.

Misaligned Resource Allocation

Another key finding is how companies allocate their AI budgets. Over half of generative AI spending currently goes to sales and marketing tools, yet MIT discovered the highest ROI in back-office automation. Eliminating outsourcing, reducing agency costs, and streamlining workflows were cited as areas where AI adoption showed the strongest business impact.

What Separates Successful AI Deployments?

According to the report, how companies acquire AI tools is critical. Purchasing solutions from specialized vendors and forming partnerships succeeded in about 67% of cases. By contrast, internal builds delivered positive results only one-third of the time.

This insight is particularly relevant in financial services and highly regulated sectors, where many firms are building proprietary generative AI systems. MIT’s data suggests, however, that companies relying solely on in-house projects face a much higher failure rate.

Challapally added that many enterprises were reluctant to reveal failure statistics. “Almost everywhere we went, enterprises were trying to build their own tool,” he said. Yet the research makes clear that externally developed tools generally outperformed.

Other success factors include giving line managers—not just centralized AI teams—the authority to drive adoption and choosing platforms capable of deep integration and long-term adaptability.

Workforce Disruption and Shadow AI

Generative AI is already reshaping workforces, particularly in customer support and administrative functions. Instead of mass layoffs, many companies are quietly reducing headcounts by not replacing employees when positions become vacant. These shifts are most evident in jobs that had already been outsourced or considered low-value.

The research also revealed the widespread use of shadow AI, where employees adopt unsanctioned tools like ChatGPT for work tasks. This poses both opportunities and risks, as organizations continue struggling to measure productivity and profit gains from such usage.

The Next Phase: Agentic AI Systems

Looking forward, advanced firms are beginning to experiment with “agentic AI”—systems capable of learning, remembering, and taking actions independently within set boundaries. This next wave could represent a major shift in enterprise AI, offering deeper automation and decision-making power across industries.

Leadership Moves

• Michael A. Discenza has been appointed VP and CFO of The Timken Company (NYSE: TKR). With 25 years of service, including a decade as VP of Finance, he steps into the role with deep institutional experience.
• John Cole has been named CFO of ELB Learning, a provider of immersive learning solutions. With over 25 years in finance and operations for Fortune 100 and 500 companies, he will focus on strengthening infrastructure for the company’s growth.

Big Deal: Cybersecurity Risks in Manufacturing

Modern manufacturing, heavily reliant on connected devices and industrial control systems, faces rising cybersecurity threats. According to Rockwell Automation’s State of Smart Manufacturing Report, many manufacturers are turning to AI-driven solutions for protection.

The survey, spanning 1,500 leaders across 17 countries, found cybersecurity is now the second-biggest external risk after inflation and economic growth. Roughly one-third of respondents manage both IT and OT cybersecurity responsibilities.

Nearly half of cybersecurity professionals (48%) said securing converged architectures will be vital over the next five years. However, challenges such as talent shortages, lack of training, and rising labor costs remain significant. As the next generation enters the workforce, cybersecurity and analytical skills are becoming key hiring priorities—showing the need to align innovation with human capital.

Going Deeper: The Cost of Losing Black Women in the Workforce

In a recent opinion column for Fortune, Katica Roy, CEO and founder of Pipeline, highlighted the economic consequences of nearly 300,000 Black women leaving the workforce in 2025.

“This isn’t a seasonal fluctuation or statistical footnote. It’s a strategic failure with long-term consequences,” Roy wrote. She stressed that Black women have historically maintained the highest labor force participation rates among all groups of women in the U.S., playing a pivotal role in industries, family incomes, and innovation.

Losing this segment, she argued, is not just a social issue but a direct threat to corporate succession planning and overall U.S. economic strength.

Overheard

“Every single Monday was called ‘AI Monday.’ You couldn’t have customer calls, you couldn’t work on budgets, you had to only work on AI projects.”

—Eric Vaughan, CEO of enterprise software firm IgniteTech, told Fortune he declared Mondays as AI-only workdays. Convinced generative AI was an existential shift, Vaughan eventually replaced nearly 80% of staff within a year when adoption lagged, according to headcount data reviewed by the publication.