The news
SAS research indicates that companies investing in trustworthy AI practices are far more likely to realize strong returns from the technology. The same study finds that trust remains a persistent barrier. Ninety-seven percent of users override AI recommendations at least some of the time. Employees show even lower trust toward increasingly autonomous AI agents.
Context
The findings come from research conducted by SAS and discussed on Bloomberg Television. Jay Upchurch, the company’s executive vice president and chief information officer, addressed the results in an interview with Bloomberg’s Anna Edwards. The work highlights a gap between AI spending and measurable outcomes that many organizations have observed in recent years. Without mechanisms that build user confidence, deployments often stall before they produce value.
The research ties specific practices to financial results. Firms that implement verifiable data governance, clear model explanations, and human oversight report higher returns than peers that skip these steps. The 97 percent override rate applies across user groups and suggests that current systems still require frequent human correction. Trust declines further when agents operate with greater independence, according to the study. Upchurch noted during the segment that these patterns appear consistently across industries that SAS tracks.
No competing data sets were presented in the segment. The single source attributes the ROI advantage directly to investments in trust-related controls rather than to model scale or compute spend.
Why it matters
For teams deploying AI inside existing products or internal tools, the SAS numbers point to a practical constraint. Override rates at this level consume the time savings that justify the original investment. Engineers who treat explainability and audit trails as core requirements rather than optional features will see fewer interventions and higher net productivity. Organizations that continue to ship opaque agents without those controls will keep paying for models that users ignore. The research does not claim that every trust measure pays off equally, but it shows a clear correlation between deliberate trust work and realized returns.
The pattern holds implications for how engineering teams structure AI projects from the start. When governance, explanations, and oversight are built into the initial design, the data from the SAS study suggests lower rates of manual correction later. This reduces the hidden labor cost that arises when staff must review or adjust outputs repeatedly. In contrast, teams that focus first on model performance and add controls afterward often encounter the high override rates the research describes.
The findings also affect decisions about autonomous agents. Lower trust in systems that act with greater independence means organizations may need additional layers of verification before scaling those agents into production workflows. Without that step, the productivity gains projected in planning documents may not appear in actual operations.
The research points to a measurable difference in outcomes tied to these choices. Companies that treat trust practices as part of the core build process rather than an add-on show stronger financial returns on their AI spend. Teams that overlook this step continue to fund systems that users bypass, which erodes the original business case.
---
Sources:
{
"word_count": 612,
"sources_used": 1,
"expanded_sections": ["context", "detail", "why_it_matters"]
}
No comments yet