Only 5% of Enterprise GenAI Deployments Show ROI as Data Trust Falters

In 2025, companies invested an estimated $30 billion to $40 billion in generative AI. [1] However, a study by MIT’s Project NANDA examined over 300 enterprise deployments and found that only 5 percent produced measurable financial returns. [1] Generative AI raises the stakes for data quality because the technology can turn questionable data into a confident answer. [1] Instead of the traditional "garbage in, garbage out" paradigm, the current dynamic is "garbage in, disaster out." [1]

Agents can now act on whatever data they encounter, and without knowledgeable human oversight, flawed results can scale across organizational operations. [1] Skipped governance allows bad or poorly understood data to become authoritative-sounding answers through chatbots. [1] If an internal chatbot provides a confident but wrong answer, the workforce may lose trust in the system. [1] To prevent this, context about where data originated, what it means, and who owns it must travel with the data as a foundation for large language models. [1]

The gap between the $30 billion to $40 billion invested and the 5 percent of deployments showing measurable returns highlights a severe ROI bottleneck driven by data governance. As the MIT Project NANDA findings suggest, the shift from human-reviewed data analysis to autonomous agents acting on unstructured data magnifies the risks of unvetted information. If organizations fail to attach origin and ownership context to their data, they risk deploying models that scale confident but incorrect decisions across their operations. What the sources don't address: How exactly the 5 percent of successful deployments structured their data governance frameworks to achieve those measurable returns.