The GenAI Gap:
The State of Artificial Intelligence in Business 2025

Of the numerous pilot programs, only about five percent generate significant growth. The overwhelming majority stagnate—showing little to no measurable impact on revenue or profit. The study is based on 150 executive interviews, a survey of 350 employees, and an analysis of 300 public AI projects. The result: a stark divide between the few success stories and the many stalled initiatives.

Aditya Challapally, lead author of the study and researcher in the NANDA project, explains: “Some pilot projects by large corporations, in collaboration with young startups, are achieving remarkable results with generative AI. They focus on a specific problem, execute with discipline, and form smart partnerships,” says Challapally.

For the remaining 95 percent of companies, the outlook is far less promising. The issue is not model quality, but a “learning gap”—both in the tools themselves and within organizations. While executives often point to regulation or technical limitations, MIT identifies the real weakness in organizational integration. Standard tools such as ChatGPT work well for individuals but fail at the enterprise level because they cannot adapt to existing workflows or integrate into company systems.

Another challenge lies in investment priorities: more than half of all generative AI budgets flow into sales and marketing solutions. Yet the MIT researchers found the greatest value in back-office applications—such as process automation, reducing reliance on external service providers, and creating more efficient operations.

The success gap is especially clear when comparing implementation strategies: purchasing specialized tools and partnering with focused startups succeeds in two-thirds of cases, while in-house development works only one-third of the time. In highly regulated industries such as financial services, many firms insist on building their own systems—yet MIT’s data shows these efforts fail more often.

“Almost everywhere we went, companies were trying to build their own tools,” Challapally notes. “But the more reliable results came from purchased solutions.” Another key factor for success: not relying solely on central AI departments, but also empowering line managers to drive adoption across the business.

AI adoption is already reshaping the workplace—particularly in customer service and administrative roles. Instead of mass layoffs, companies are increasingly choosing not to backfill positions when employees leave. The functions most affected are those that have traditionally been outsourced.

Equally important is equipping end users to adapt to this evolving workplace, ensuring they can work effectively with the new systems and remain fully engaged in AI projects.

The report also highlights the widespread use of “shadow AI”—unauthorized tools such as ChatGPT that employees use on their own. At the same time, companies struggle to precisely measure AI’s impact on productivity and profitability. Risks remain, including potential data leaks and even contract violations.

Looking ahead, the most advanced companies are already experimenting with AI agent systems that can learn, remember, and act independently within defined boundaries. These systems may well mark the beginning of the next phase of AI adoption in business.