MIT NANDA research finds that only 5% of organizations successfully scale AI tools into production. American companies have invested between $35 billion and $40 billion in generative AI projects, yet so far, they have seen almost no return on investment. According to a report from MIT NANDA (Network AI Agents and Decentralized AI), 95% of corporate organizations have received zero returns from their AI investments. Only 5% of organizations have successfully integrated AI tools into production at scale. The report is based on structured interviews with 52 corporate leaders, an analysis of over 300 public AI projects and announcements, and a survey of 153 business professionals. The authors of the report—Aditya Challapally, Chris Pease, Ramesh Raskar, and Pradyumna Chari—attribute this "generative AI gap" to the inability of AI systems to retain data, adapt to environments, and learn continuously, rather than a lack of infrastructure, learning resources, or talent. > The "generative AI gap" is most evident in deployment rates, with only 5% of customized enterprise AI tools making it into production. "The 'generative AI gap' is most evident in deployment rates, with only 5% of customized enterprise AI tools making it into production," the report states. "Chatbots succeed because they are easy to try and flexible, but they fail in critical workflows due to a lack of memory and customization capabilities." As one anonymous CIO said in an interview with the authors, "This year we saw dozens of demos. Maybe only one or two were truly useful. The rest were either 'shelled' products or science experiment projects." The authors' findings align with other recent studies indicating that corporate leadership's confidence in AI projects is declining. The NANDA report does mention that a small number of companies have found uses for generative AI, and that the technology is having a substantial impact in two of nine industrial sectors—technology and media & telecommunications. For the remaining sectors—professional services, healthcare & pharmaceuticals, consumer & retail, financial services, advanced manufacturing, and energy & materials—generative AI has remained irrelevant. The report cites an anonymous COO from a mid-market manufacturing company: "The hype on LinkedIn is overwhelming, saying everything has changed, but in our actual operations, there has been no fundamental change. We process some contracts faster, but that's about it." One thing that is indeed changing is the employment landscape, at least in the affected industries. The report notes that in technology and media, "over 80% of executives expect to reduce hiring within 24 months." According to the authors, layoffs driven by generative AI are primarily occurring in non-core business activities that are often outsourced, such as customer support, administrative processing, and standardized development tasks. "These positions were already vulnerable due to their outsourced nature and process standardization before AI implementation," the report states, noting that 5% to 20% of support and administrative processing roles in affected industries have been impacted. According to The Register, Oracle's recent layoffs reflect its efforts to balance AI capital expenditures, which have become a heavy burden on the necks of American tech giants. Meanwhile, at IBM, employees feel that AI has been used as an excuse to shift jobs overseas. Regardless of the public reasons and true motivations behind the layoffs, generative AI is indeed impacting the technology and media & telecommunications sectors, which are also the areas where it is most widely adopted. Although about 50% of AI budgets are allocated to marketing and sales, the report's authors suggest that corporate investments should flow toward activities that can produce meaningful business outcomes. This includes front-end lead qualification and customer retention, as well as back-end reductions in business process outsourcing, advertising agency spending, and financial services risk assessments. The report points out that general tools like OpenAI's ChatGPT perform better than customized enterprise tools, even if those enterprise tools use the same underlying AI models. The rationale presented in the report is that employees are often more familiar with the ChatGPT interface, leading to more frequent use—this is a result of employees' spontaneous "shadow IT". The report quotes an interview with a company lawyer who described her mid-sized law firm's dissatisfaction with a specialized contract analysis tool that cost $50,000. "The summaries provided by the AI tool we purchased were very rigid, and the customization options were limited," the lawyer told researchers. "With ChatGPT, I can guide the conversation, iterate repeatedly, until I get exactly what I need. The fundamental quality difference is obvious; ChatGPT consistently produces better results, even though our vendor claims they use the same underlying technology." The authors believe that companies that successfully cross the "generative AI gap" approach AI procurement more like outsourcing business process services rather than as customers of software as a service (SaaS). "They demand deep customization, drive applications from the front lines, and hold vendors accountable for business metrics," the report concludes. "The most successful buyers understand that crossing this gap requires building partnerships, not just purchasing products."®