Companies are embracing artificial intelligence with high expectations, but the reality is harsh. A new MIT report titled The GenAI Divide: State of AI in Business 2025 reveals that as many as 95% of generative AI pilot programs in businesses fail to deliver rapid revenue growth. Instead, most of them stall and have no visible impact on financial performance. The report is based on 150 interviews with executives, a survey of 350 employees, and an analysis of 300 public AI deployments. Sheryl Estrada, the author of an article in Fortune magazine, discussed the topic with Aditya Challapally, the report’s lead author and a researcher with MIT’s NANDA project.
Aditya Challapally explains that some large companies and young generative AI startups are indeed excelling. For example, startups led by nineteen- or twenty-year-old founders have managed to increase their revenue from zero to $20 million within a single year. They do this by focusing on a single problem, addressing it thoroughly, and collaborating with companies that use their tools. But for 95% of the companies in the report’s data, generative AI implementation is failing. The main problem is not the quality of AI models, but a learning gap affecting both the tools and the organizations. Executives often blame regulations or model performance, but MIT’s research points to flawed integration into business processes.
Generic tools such as ChatGPT work very well for individuals because of their flexibility, but they fail in businesses because they do not learn from or adapt to workflows, according to Aditya Challapally. The data also reveals a mismatch in resource allocation. More than half of generative AI budgets go toward sales and marketing tools, yet the greatest return on investment comes from back-office automation—that is, eliminating business process outsourcing, reducing spending on external agencies, and streamlining operations.
What Determines the Success of AI Deployment?
The way companies adopt AI plays a crucial role. Purchasing AI tools from specialized vendors and building partnerships succeeds in about 67% of cases, while in-house development succeeds only one-third of the time. This finding is particularly important in regulated industries such as financial services, where many companies are building their own proprietary generative AI systems in 2025. However, MIT’s research suggests that going it alone leads to far more failures.
Companies in the survey were often reluctant to share their failure rates, as Aditya Challapally noted. Nearly everywhere the researchers looked, businesses were trying to create their own tools, but the data shows that purchased solutions deliver more reliable results. Other important factors for success include empowering lower-level managers—not just central AI labs—to drive adoption, and selecting tools that integrate deeply and adapt over time.
Changes in the labor market are already underway, particularly in customer support and administrative roles. Rather than carrying out mass layoffs, companies are more likely to leave vacant positions unfilled. Most of the changes are concentrated in jobs that were previously outsourced because of their low value. The report also highlights the widespread use of "shadow AI"—unauthorized tools such as ChatGPT—and the persistent challenge of measuring AI’s impact on productivity and profits.
The Future of AI in Business
The most advanced organizations are already testing agentic AI systems that learn, remember, and act independently within defined boundaries. This indicates what the next phase of AI in business might look like. The MIT report emphasizes that successful cases lead to significant transformations, such as rapid revenue growth at startups, but most companies need to rethink their approach to avoid failure.
