Share with your CDO
A new SAS report, researched with IDC across 2,699 decision-makers in 28 countries, makes a stark quantitative case that AI governance isn’t overhead, it’s alpha. Organizations scoring highest on trustworthy AI practices, defined across data quality, model oversight, explainability, responsible policy, and audit accountability, were 15 times more likely to report strong ROI on AI projects (62% versus 4%). Only 17.5% of enterprises have data infrastructure mature enough for agentic AI today. The full findings are in the SAS Data and AI Impact Report.
What this means for your business
The 15x ROI gap is the kind of number that sounds like vendor hyperbole until you read the mechanism. Nearly all users (97.2%) override AI recommendations at least sometimes, and the top reason isn’t inaccuracy, it’s that the system can’t explain its reasoning. Override behavior compounds silently: every manual correction tells the organization that the AI isn’t trustworthy, which depresses adoption, which starves the model of feedback, which keeps it from improving. If your organization has deployed AI broadly but ROI conversations keep stalling, the override rate is probably the diagnostic you’re not pulling.
SAS has obvious incentives to publish a study where “trustworthy AI” practices (the exact category its governance platform sells into) predict financial outperformance, and the survey’s industry scope, banking, insurance, life sciences, and public sector, skews toward regulated verticals where governance spending is already mandated. That selective framing likely inflates the measured ROI gap for the broader enterprise market. Even so, the directional finding is hard to dismiss: the report’s leaders aren’t deploying different AI technology than the laggards, they’re managing it differently. That distinction shifts accountability squarely to data and AI leadership rather than to the vendor stack.
The more consequential number buried in the report is the infrastructure gap: only 17.5% of enterprises have data infrastructure ready for agentic AI, the category every major vendor is currently pitching as the next deployment wave. Organizations in that 82.5% majority aren’t facing a governance maturity gap so much as a data plumbing gap, and governance frameworks built on top of weak data infrastructure produce audit trails for unreliable outputs. The budget question to reweigh isn’t whether to invest in AI governance tooling, it’s whether that spending runs ahead of or behind the data foundation work it depends on.
Concept deep-dive: Explainability
Explainability refers to an AI system’s ability to show, in terms a human reviewer can evaluate, why it produced a specific output or recommendation. It exists because complex models, particularly neural networks, generate predictions through patterns across thousands of variables with no single readable decision path. Think of it as the difference between a colleague saying “here’s my answer” and “here’s my answer, and here’s the chain of evidence behind it.” Without it, human oversight becomes theater, and the SAS data suggests employees know it.
Based on reporting from SAS study links trustworthy AI practices to higher enterprise ROI, originally published 2026-09-14 15:29:00.
