AI assistants: what should they optimise in Lausanne?
In SHR — Swiss Human Resources’ programme “Créer ses assistants et agents IA sans coder” (Create your own AI assistants and agents without coding), choosing an objective deserves as much attention as building the system. The management question is precise: how can an assistant tasked with speeding up request handling be prevented from favouring closure over resolution?
What the literature establishes
Steven Kerr (1975, Academy of Management Journal) analyses the gap between the behaviours organisations reward and those they say they want. Drawing on organisational examples, his article shows why rewarding an easily observable outcome can divert attention from a less directly measurable purpose. For an AI assistant, this means scrutinising the chosen objective: producing a reply, closing a request and solving a problem are not equivalent.
When measurement changes the work
Wendy Espeland and Michael Sauder (2007, American Journal of Sociology) examine how rankings transform the practices of US law schools. They describe, in particular, how numerical comparisons and the expectations generated by rankings redirect organisational behaviour. The relevant mechanism here concerns those managing the system: a dashboard focused on closed requests can shape their configuration and evaluation choices.
The premature conclusion
One might conclude that metrics should be abandoned, or simply that more should be added. These studies establish neither prescription: they invite us to examine which behaviours measurement makes advantageous. A closure rate remains useful, provided it is not treated as sufficient evidence of resolution. (our executive and employee training programmes)
What these studies cannot establish
These articles address neither generative models nor no-code agent-building tools. They therefore do not demonstrate that an assistant will spontaneously game a metric: a measure used in a dashboard does not, by itself, become a technical objective of the system. The application concerns human decisions about design, selection and deployment first; their effects must then be checked in practice.
A practical check in Lausanne
In Lausanne, a medtech team, a university service, an international sports federation or a Lake Geneva region scale-up could test an assistant on non-sensitive internal requests, using a resolution criterion defined before the trial. Over a predefined period, record the proportion of closed requests subsequently reopened for the same need, keeping the follow-up duration identical for every request. Reviewing the cases helps distinguish an incomplete response from a new need; faster closure accompanied by more reopenings would signal a trade-off to investigate, not an established improvement. Within the SHR programme, this check can serve as a design exercise: define business success before configuring the assistant. To go further: explore the Building AI Assistants and Agents Without Coding training in Lausanne in the canton of Vaud, or browse our executive and employee training programmes in Switzerland.
In pictures: Building AI Assistants and Agents Without Coding in Lausanne
- building ai assistants and agents without coding
- Lausanne
- research
- training lausanne