AI assistants: standardising without making service rigid
Building a no-code assistant makes it possible to turn a procedure into a guided workflow quickly. The management question is specific: how can requests be handled more consistently without preventing appropriate treatment of situations that fall outside the standard process?
What the literature establishes
Paul S. Adler and Bryan Borys (1996, Administrative Science Quarterly) distinguish two approaches to formalisation: one helps people master their work, while the other primarily seeks to secure compliance. In their theoretical framework, a procedure can support work when it makes its operation understandable and allows people to address the difficulties they encounter. Applied to AI assistants, this distinction shifts the design criterion: getting people to follow the steps is not enough to demonstrate that the tool helps them work.
A routine is more than its instructions
Martha S. Feldman and Brian T. Pentland (2003, Administrative Science Quarterly) distinguish the general representation of a routine from its concrete enactments by people in particular situations. Their theoretical analysis explains how routines can generate stability, but also flexibility and change. For an assistant, the risk would be to treat the documented procedure as an exhaustive representation of actual work.
The premature conclusion
One might conclude that every user should be free to bypass the proposed workflow. These studies do not justify that rule: a useful adaptation is not equivalent to abandoning a safety requirement or a regulatory obligation. A more cautious design implication is to distinguish non-negotiable requirements, adaptable steps and exceptions that must be referred to a qualified person. (our executive and employee training programmes)
What these studies do not demonstrate
Both articles are theoretical contributions to organisational research, not evaluations of generative assistants or no-code platforms. They therefore do not establish that adding an “exceptional situation” button will, by itself, improve service quality. Simply counting completed cases would nevertheless be insufficient to test this: a workflow can be completed even though a request has been misclassified or a difficulty has been shifted elsewhere.
A practical check in Sion
As part of SHR's “Creating your own AI assistants and agents without coding” programme, an exercise in Sion could address the routing of administrative requests relating to hydropower, healthcare, viticulture and Alpine tourism, without entrusting sensitive decisions to the assistant. Before testing, professionals would classify fictional cases as standard or exceptional and define their expected destination. Comparing a fixed workflow with one offering an exception route would make it possible to measure the proportion of exceptional cases correctly referred, retaining each case's initial classification and observed outcome. Standard requests transferred unnecessarily should also be recorded: the aim is better routing, not referring everything to a person. To go further: explore the Building AI Assistants and Agents Without Coding training in Sion in the canton of Valais, or browse our executive and employee training programmes in Switzerland.
In pictures: Building AI Assistants and Agents Without Coding in Sion
- building ai assistants and agents without coding
- Sion
- research
- training sion