AI adoption: who checks what the team accepts?

A well-written answer can enter a deliverable without anyone explicitly deciding to trust it. The managerial question is specific: how should verification responsibilities be allocated so that the team does not confuse using AI with delegating judgement?

What the literature establishes

John D. Lee and Katrina A. See (2004, Human Factors) review research on trust in automation: the objective is not maximum trust, but reliance appropriate to the system’s capabilities and the context. Trust helps guide reliance on a tool particularly when its operation cannot be fully understood. For managers, this suggests distinguishing willingness to use AI from the appropriateness of relying on a particular answer.

The mechanism: monitoring can weaken

Raja Parasuraman and Victor Riley (1997, Human Factors) distinguish the use, misuse, disuse and abuse of automation. Their analysis describes, among other things, how excessive reliance can be accompanied by insufficient monitoring of results. Applied to AI, this framework prompts an examination not only of what the tool produces, but also of what the team stops checking.

The premature conclusion

One might conclude that everything should be checked systematically. These studies do not, however, establish that exhaustive checking is the best arrangement for every task. A more cautious managerial implication is to make verification proportionate to the consequences of an error, its detectability and the possibility of reversing the decision. (our executive and employee training programmes)

What the evidence cannot determine

These two articles are reviews of automation research, not trials of today’s generative assistants in Geneva-based businesses. They illuminate a supervision problem without supplying a universal checking rate or guaranteeing that human approval will detect errors. An instruction to “have someone review it” therefore remains inadequate unless it specifies what to check, which reference sources to use and who is responsible.

A practical check in Genève

Within SHR — Swiss Human Resources’ programme « Conduire l’adoption de l’IA dans son équipe », one exercise is to define a critical check for an authorised use: a client-facing statement in private banking, a documentary reference in an international organisation, a product specification in luxury watchmaking or a contractual clause in commodity trading. Over two weeks, the team can record the relevant deliverables, the required check, the reference consulted and the person who approved them, without copying sensitive data. The proposed indicator is the proportion of these deliverables with a documented check before release, supplemented by verification time and corrections made. This record measures whether checking took place, not the absence of errors; it nevertheless makes it possible to establish whether stated responsibility translates into actual practice. To go further: explore the Leading AI Adoption in Your Team training in Geneva, or browse our executive and employee training programmes in Switzerland.

In pictures: Leading AI Adoption in Your Team in Geneva

• leading ai adoption in your team
• Geneva
• research
• training geneva