A second edition of the Data and AI Impact Report: The New Economics of Trust, produced by SAS and IDC, examines which practical practices are linked to successful AI investments. The study finds that organizations actively applying trustworthy AI principles are far more likely to report strong returns from AI projects: 62% of such organizations reported strong or high ROI, compared with just 4% of lagging organizations — roughly a 15-fold difference.
Components of trustworthy AI and their impact
According to the report, a relatively small group of companies that follow the strictest practices in governance, data quality and auditability consistently outperform peers. These organizations achieve at least twice the ROI on their AI developments compared with others; fewer than 5% of organizations that fall behind on trustworthy AI can claim similar results.
Bryan Harris, Chief Technology Officer at SAS, said: “When AI works well, it can have incredible impact.” He also warned that error rates for advanced AI agents on complex tasks can exceed 25%, which is unacceptable where stakes are high. Harris argued that domain expertise must be embedded into agent workflows and that human control and oversight must remain in place to ensure accuracy and reliable repeatability.
Chris Marshall, Vice President at IDC, added: “As AI becomes more autonomous, organizations face a new challenge: trusting systems they do not fully understand.” Marshall noted that stronger oversight, explainability, accountability and robust data foundations are increasingly preconditions for successfully scaling AI.
Explainability and the cost of overrides
The research highlights that employees in many organizations are increasingly reluctant to rely on systems that cannot explain their outputs. The more autonomous an AI system, the greater this risk, making explainability a key success factor. Employees frequently override AI decisions and make manual corrections; this deepens the trust gap and costs organizations time, productivity and money. Because AI decisions are only as good as the data they rely on, reducing overrides requires first building solid data foundations.
Key findings include:
- 97.2% of users override AI recommendations at least occasionally.
- Employees most often override AI — regardless of whether they considered the result correct — because the system could not justify its decision.
- Trust in generative AI stands at 76%, while trust in agentic AI (AI agents) is 66% — indicating growing concern as systems become more autonomous.
Trustworthy AI drives business success
The report shows a widening ROI gap between organizations that prioritize trustworthy AI and those that do not. Organizations investing in trustworthy AI are 15 times more likely to report strong or high returns (62% vs. 4%). The companies following the most trustworthy practices achieve on average 1.85 times greater gains across 13 business metrics, including revenue growth, cost savings and customer experience.
Additionally, 85% of these leading organizations plan to increase investments in trustworthy AI by more than 10% this year, further widening their advantage.
Poor data foundations waste time and money
Many organizations still build AI on immature or outdated data and infrastructure. The report finds only 17.5% have a fully optimized data infrastructure suitable for AI agents’ needs. Organizations with an optimized data foundation are four times more likely to expect strong ROI from AI projects and six times more likely to have mandatory data quality and explainability controls required for trust.
Industry results and methodology
The findings are based on a survey of 2,699 decision-makers who have visibility into or influence over their organizations’ data and AI initiatives. The survey covered 28 countries and focused on four industries: banking, insurance, life sciences and the public sector. Notable industry takeaways:
- Leading banks go beyond regulatory compliance: 85% of leading banks have a governance framework for AI, compared with 29% of lagging banks.
- 41% of leading public sector organizations plan to increase trustworthy AI investments by more than 20% next year — matching the most ambitious firms in any industry.
- 23% of life sciences companies have deployed AI across the organization, the highest adoption rate among the sectors studied.
What constitutes trustworthy AI?
The report defines trustworthy AI as systems designed to operate reliably, ethically and safely, comply with regulations, and clearly demonstrate how decisions were reached. Users and decision-makers across the organization must be able to hold AI systems accountable for faulty or missing outcomes through predefined lines of responsibility. All AI systems should be subject to governance principles and demonstrable compliance with clear rules.
How leaders are identified
Organizations were scored on a 100-point scale across five dimensions of trustworthy AI: (1) data quality and data governance; (2) model governance and oversight; (3) explainability and fairness; (4) responsible AI policy; and (5) auditability and accountability. The report considers organizations leaders when they score at least 80 points on average.
The report underscores that investing in trust is not merely a compliance exercise but a source of competitive advantage — provided organizations build robust data foundations, clear governance and explainability into their AI deployments.
Image credits: SAS
