AI has leapt forward — but data practices lag

Four years have reshaped the AI landscape. Since the first edition of this study in 2021, generative AI has accelerated capabilities across modalities: models now commonly handle text, audio, video and other unstructured formats, and autonomous AI agents are beginning to act and reason with less human supervision.

Data quality still determines outcomes

Despite rapid progress in model capabilities, the study underscores a perennial truth: model outputs are only as good as the data that feeds them. Data management tools and best practices have improved, but according to the second edition of this report most organizations are not adopting those advances quickly enough to match AI development. As a result, many companies still fail to translate AI experiments into measurable business results.

Survey and interviews: how the findings were gathered

MIT Technology Review Insights surveyed 800 senior data and technology executives to assess organizational data performance amid the rise of generative AI. The research also included in-depth interviews with 15 technology and business leaders to add context and qualitative insight.

Main findings

• Few data teams are keeping pace with AI. The percentage of organizations that consider themselves data 'high achievers' has not increased since 2021: 12% in 2025 versus 13% in 2021. Challenges include shortages of skilled talent, difficulty accessing fresh data, tracing data lineage, and managing security complexity — all critical for AI success.
• AI is not yet delivering widespread business impact. Only 2% of surveyed senior executives rate their organizations highly for delivering measurable business results from AI. While two thirds of respondents have deployed generative AI in some form, only 7% report wide deployment.

What this means for organizations

The report suggests that technical advances alone are not enough. Companies that want AI to drive business outcomes must invest in data access, governance, lineage and security at the same pace they invest in models and experimentation. Without modernized data practices, even the most capable AI systems will underdeliver.

This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review's editorial staff. The report was researched, designed, and written by human writers, editors, analysts, and illustrators; any AI tools used were limited to secondary production processes and passed thorough human review.