Itdaily - Nearly a quarter of AI recommendations in the Benelux are adjusted or ignored

Nearly a quarter of AI recommendations in the Benelux are adjusted or ignored

Cartoon of a frowning man and a white robot working side by side on laptops.

Employees in the Benelux adjust or completely ignore nearly a quarter of AI recommendations. Despite improved AI governance, the business impact remains lagging.

Employees in the Benelux deviate from recommendations made by AI systems in an average of 24 percent of cases. This is according to the second Data and AI Impact Report: The New Economics of Trust by SAS, prepared with insights from IDC. The results reveal a tension between growing investments in trustworthy AI and a business impact that is barely keeping pace.

According to the report, a hidden cost arises when employees cannot trust or understand AI outcomes. They then have to check, justify, or manually correct recommendations, causing part of the expected time and productivity gains to disappear.

Explainability

Nearly half of the interventions in AI recommendations (49.2%) occur because the output provides insufficient explanation or transparency to be trusted. Potential bias is the reason in 21.3 percent of interventions, while factual errors account for only 9.8 percent. Employees therefore primarily encounter a lack of context, rather than necessarily blatant incorrect results.

The Trustworthiness Index for Benelux organizations rose from 38.5 to 55.6 points in one year. This index measures, among other things, data quality, model governance, explainability, and accountability. However, the Impact Index, which maps the business impact of AI, remained virtually flat and dropped slightly from 47.5 to 46.9 points. Better governance does not yet automatically translate into noticeable added value today.

Governance grows, impact lags behind

According to the report, organizations in the Benelux have taken clear steps regarding data quality, oversight, explainability, and accountability, but the region remains below global benchmarks. Only 4.6 percent of organizations mandate the use of tools that explain AI outcomes for every project. This is at odds with the fact that insufficient explanation is the main reason for ignoring AI recommendations.

“When AI works, the impact can be enormous,” says Bryan Harris, CTO at SAS. “At the same time, it has been shown that advanced agents can have error rates of more than 25 percent for complex tasks. For important decisions, that is unacceptable. To achieve accuracy and repeatability, organizations must integrate domain expertise into agentic workflows and keep humans at the center of governance and oversight. Organizations that succeed in this can narrow the trust gap and achieve a competitive advantage with AI.”