Defining the boundaries of AI:
Why some decisions remain a human responsibility
Amsterdam – No matter how autonomous AI systems, such as ERP systems, become, there are decisions that managers are not willing to hand over to AI. More than half (53%) believe AI should never be allowed to independently enter into contracts with suppliers or customers, while 51% want dismissal decisions to remain with a human. Meanwhile, 45% believe salaries and employment conditions should never be determined by AI. These findings come from research by Dynamic People among 317 IT and business decision-makers at medium-sized and large organizations in the Netherlands.
Step 1: Define where AI can act autonomouslyand where humans stay in control,
Step 2: Make explainability a requirement for autonomous decisions,
Step 3: Set clear ownership, thresholds and intervention points.
Step 1: Define where AI can act autonomouslyand where humans stay in control,
Step 2: Make explainability a requirement for autonomous decisions,
Step 3: Set clear ownership, thresholds and intervention points.
Organizations also want to retain control over financial decisions. Six in ten (61%) believe AI should never be allowed to independently make financial decisions above a certain amount. Exactly where that threshold should lie varies considerably. The most frequently mentioned thresholds are €10,000 (29%) and €50,000 (22%), while 18% believe AI should never be allowed to make financial decisions independently at all.
When organizations do allow AI to act autonomously, there are clear moments at which they say they will intervene manually. Nearly two-thirds (64%) do so as soon as an AI system exceeds a financial threshold, while six in ten (60%) intervene when AI makes high-risk decisions involving customers, suppliers or contracts. For 36%, intervention only takes place afterwards, once a decision has had a negative outcome.
Yet human intervention does not always happen at the right moment. Nearly eight in ten (79%) say human intervention often only takes place after the damage has already been done. Earlier intervention is only possible when it is clear why a system has made a particular decision. A quarter (26%) therefore intervene as soon as AI is unable to explain its decision. Explainability also plays a role in how far organizations are willing to go with autonomy: 83% believe full autonomy will remain an illusion as long as AI decisions need to be explainable.
This research was conducted as a quantitative online study commissioned by Dynamic People. The target group consisted of decision-makers responsible for IT, ERP and CRM solutions and/or financial decisions within organizations with 50 or more FTEs, operating in business services, wholesale and distribution, financial services or commodity trading. A total of 317 respondents participated in the study. The fieldwork was conducted by Markteffect between February 24 and March 9, 2026.