IA on the job: will a generated plan actually be carried out?
With generative AI, obtaining a list of actions is becoming easier than securing the conditions needed to carry them out. The useful management question is therefore specific: how can an AI-proposed action become behaviour that is actually triggered at the right moment?
What the literature establishes
Peter M. Gollwitzer (1999, American Psychologist) distinguishes an intention to pursue a goal from an implementation intention, which links an identifiable situation to a specified response. His review explains why wanting to complete a task is not always enough to start it. For a plan developed with AI, the relevant distinction is therefore not between a short list and a detailed one, but between a desired outcome and an explicitly recognised opportunity to act.
The mechanism: linking a situation to an action
Peter M. Gollwitzer and Veronika Brandstätter (1997, Journal of Personality and Social Psychology) investigated how specifying when and where to act can support goal attainment. Their research supports preparing the trigger for action rather than merely restating the goal. “When the missing document arrives in the file, I check its validity before forwarding it” specifies an opportunity and a response that “complete the file” leaves undefined.
The premature conclusion
One might conclude that it is enough to ask AI to turn every recommendation into an “if… then…” rule. Yet these studies do not establish that a rule generated by an assistant will be adopted or that it will specify an appropriate action. Employees must be able to recognise the situation, accept the action and have the means to carry it out; precise wording does not resolve a disputed goal or missing authorisation. (our executive and employee training programmes)
What the evidence does not establish
These publications address neither generative AI nor regulated processes in Geneva-based businesses. They therefore cannot quantify a local improvement in reliability or productivity. In practice, an action left undone may also reflect an absent trigger, a changed priority or an external dependency: counting ticked boxes without examining these conditions conflates different situations.
A practical check in Genève
Within SHR’s “IA on the job” programme, one possible exercise is to select a recurring, authorised action: checking a document in private banking, following up on an approval in an international organisation, checking a service agreement in luxury watchmaking or verifying a delivery document in commodity trading. Over two weeks, each participant validates a rule specifying an observable trigger, the expected action and its deadline, then records trigger occurrences and actions completed within that deadline, without entering sensitive data into the assistant. The measure is the proportion of occurrences followed by the agreed action within the specified time, accompanied by reasons for non-completion. Comparison with a baseline period for the same task makes changes visible, without being sufficient to attribute any improvement to AI. To go further: explore the AI on the Job training in Geneva, or browse our executive and employee training programmes in Switzerland.
In pictures: AI on the Job in Geneva
- ai on the job
- Geneva
- research
- training geneva