AI adoption: should procedures be set in stone?
A team introducing AI must decide what to standardise without assuming that every work situation is alike. The management question is precise: how can a shared procedure be established while making situations that require adaptation visible?
What the literature establishes
Martha Feldman and Brian Pentland (2003, Administrative Science Quarterly) distinguish the general representation of a routine from its concrete performances by particular people in particular situations. Their analysis shows why a routine can be a source of stability, but also of change. Applied to AI, this distinction cautions against confusing the existence of an operating procedure with consistency in the work actually performed.
The tool alone does not determine practice
Wanda Orlikowski (1992, Organization Science) proposes a model in which technology is both shaped by human action and involved in structuring that action. Organisational conditions and patterns of use thus help explain what the same tool becomes in the workplace. For managers, this suggests that identical settings do not guarantee identical practices.
The premature conclusion
One might conclude that everyone should be free to adapt their use of AI. These studies do not justify that conclusion: recognising variation in practice does not make every variation acceptable. In a regulated activity, an adaptation may require review and approval before implementation; the task is to distinguish an exception requiring examination from a workaround that must be stopped. (our executive and employee training programmes)
What these studies cannot promise
These articles predate generative AI and primarily offer theoretical frameworks supported by organisational analyses. They establish neither the optimal degree of standardisation for an AI use case nor a causal effect on its quality. A documented procedure therefore does not, on its own, prove that a process is under control; conversely, an observed deviation does not prove that the procedure is flawed.
A practical check in Basel
As part of SHR’s programme « Conduire l'adoption de l'IA dans son équipe », one proposed application in Basel is to track an authorised, narrowly defined use case for two weeks, such as preparing a summary from non-sensitive documents. In a pharmaceutical, life sciences, Rhine logistics or international R&D team, each case would be logged, indicating whether the procedure is sufficient or an adaptation is requested, without changing mandatory controls. The measure would be the proportion of cases requiring adaptation out of all logged cases, verifiable against the register, with the reason and the authorised manager’s decision recorded. This record does not measure AI performance: it checks whether the shared procedure covers the work encountered before its scope is extended. To go further: explore the Leading AI Adoption in Your Team training in Basel in the canton of Basel-Stadt, or browse our executive and employee training programmes in Switzerland.
In pictures: Leading AI Adoption in Your Team in Basel
- leading ai adoption in your team
- Basel
- research
- training basel