good morning. Companies are betting on AI, but almost every enterprise pilot is stuck at the starting line.
genai divide: AI status in Business 2025a new report published by MIT's Nanda Initiative reveals that while generating AI is promising for businesses, most initiatives to drive rapid revenue growth have been leveling off.
Despite the rush to integrate powerful new models, around 5% of AI pilot programs have achieved rapid revenue acceleration. The majority of stalls provide little or no measurable impact on P&L. Based on 150 interviews with leaders, a survey of 350 employees and an analysis of 300 public deployments, the study portrays a clear gap between success stories and stalled projects.
To unleash these findings, I spoke with Aditya Challapally, the lead author of the report, and research contributors at Project Nanda at MIT.
“Pilots and young startups from some large companies are really good at generative AI,” says Challapally. For example, a 19- or 20-year-old startup “we've seen revenues go from zero to $20 million a year,” he said. “That's because they're smart partners with companies that choose one problem, do it well and use their tools,” he added.
However, 95% of companies in a dataset lack the implementation of generated AI. A core issue? It's not the quality of the AI model, it's the “learning gap” between both the tool and the organization. Executives often condemn the performance of regulations and models, but MIT's research points to flawed enterprise integration. A common tool like ChatGpt is Excel for individuals for flexibility, but it stalls its use in enterprises because it does not learn or adapt from workflows, Challapally explained.
The data also reveals inconsistencies in resource allocation. While more than half of your generative AI budget is dedicated to sales and marketing tools, MIT has discovered that it is the largest ROI in back-office automation. This reduces business process outsourcing, external agency costs, and streamlined operations.
It's important to know how companies employ AI. The purchase of AI tools from specialized vendors and building partnerships have about 67% of the time successful, while internal builds are often only a third successful.
This finding is particularly relevant to financial services and other highly regulated sectors in 2025 where many companies are building their own generation AI systems. However, MIT's research suggests that companies see far more failures when going solo.
The companies surveyed were often hesitant to share failure rates, Challapally noted. “Nearly everywhere we went, companies were trying to build their own tools,” he said, but the data showed that the solutions they purchased had more reliable results.
Other key factors for success include strengthening the power of line managers to encourage recruitment and selecting tools that can be deeply integrated and adapted over time.
Workforce disruption is already underway, especially in the role of customer support and management. Rather than a massive layoff, businesses are vacant, so they are increasingly less filling up. Most changes focus on previously outsourced work due to their low value.
The report also highlights the widespread use of “Shadow AI” (a non-trusted tool such as ChatGPT) and the ongoing challenge of measuring the productivity and profit impact of AI.
Going forward, the most advanced organizations are already experimenting with agent AI systems that can learn, remember, remember and act within set boundaries.
