MIT Report: Most organizations do not see business returns on Gen AI investments
A recent report from MIT Media Lab found that despite enterprise spending on generated AI at $3-40 billion, 95% of organizations are not looking at business returns.
The author of the July 2025 report, entitled “The Genai Divide: AI Status in Business 2025,” wrote: There is no measurable P&L effect. ”
Meaning of “Gen Ai divide”
Dividing is defined by high adoption but by low transformation. The report says that only two industries show clear signs of structural disruption, while seven others say “indicating extensive experiments without change.”
This was supported by the AI market disruption index and included interview estimates from middle market manufacturing COOs.
Production from pilot: most efforts stall
The sharpest evidence of disparity is development. “Only 5% of custom enterprise AI tools reach production.” The report characterizes this as a 95% failure rate for enterprise AI solutions, resulting from vulnerable workflows, weak context learning, and inconsistency with day-to-day operations. It also recorded user skepticism about vendor offerings. “We've seen dozens of demos this year. Maybe one or two are really useful. The rest is a rapper or a science project.”
Enterprise operates the most pilots, but converts the fewest pilots. Mid-market organizations travel faster from pilots to full implementation (approximately 90 days) than large companies (9 months or more).
Number of recruits and business impact
General purpose tools are widely investigated, but their impact is limited. [ChatGPT/Copilot]and reported nearly 40% of deployments, but “these primarily increase individual productivity rather than P&L performance. Meanwhile, 60% of organizations evaluated enterprise-grade systems,” but only 5% reached the pilot stage and reached production.”
Root Cause: Learning Gap
The central explanation of the report is that the core barrier is learning rather than infrastructure, regulation, or talent. “Most Genai systems do not retain feedback, adapt to context, or improve over time.”
Many users prefer the draft consumer LLM interface, but refuse mission-critical work due to lack of memory and persistence. One interviewee explained: “It's great for brainstorming and initial drafting, but it doesn't retain the knowledge of the client's preferences or learn from previous edits. It requires repeated mistakes and extensive contextual input for each session.
This report briefly summarized this gap. “The very limitations of ChatGpt reveal the core issues behind the Gen AI disparity: forget the context, do not learn, and cannot evolve.” For complex, long-term tasks, humans remain strong preferences.
Shadow AI: Workers Intergrievably Overcome Disparities
While the official programme is behind, “Shadow AI Economy” has appeared. “Only 40% of 40% say they have purchased an official LLM subscription, but over 90% of companies have reported regular use of personal AI tools for work. This pattern shows that individuals can overcome the gap with flexible tools, even if their enterprise initiatives stall.
Why is this important?
For teams responsible for operating AI in cloud environments, the report showed that the bottleneck lies in systems that allow them to learn, remember and integrate workflow systems. “Dividing” is not about model IQ or raw infrastructure capacity, but about embedding adaptive behavior into application layers and process orchestration.
Methodology
This report is based on a multi-bill research design that was implemented between January and June 2025. The researchers conducted a systematic review of more than 300 publicly published AI initiatives, held 52 structured interviews from industry organizations, and collected 153 survey responses from senior leaders at four large meetings. Company-specific data and citations have been anonymized to comply with disclosure policies.
The report is available on Nandapapers Github Repo.
About the author
David Ramel is the editor and writer of Converge 360.
