AI assistants: is a volunteer pilot enough in Berne?
Building an assistant without coding makes it possible to put a prototype into interested colleagues' hands quickly. The management question is more specific: do results obtained with these volunteers justify deployment across the whole team?
What the literature establishes
James Heckman (1979, Econometrica) formalises a selection problem: when entry into a sample depends on characteristics related to the outcome being studied, observations can give a biased picture of the target population. Applied to an AI assistant pilot, this framework prompts us to examine who agrees to participate, not just what participants accomplish. It does not demonstrate that volunteers systematically perform better, but it rules out taking their representativeness for granted.
Acceptance is not a performance measure
Fred Davis (1989, MIS Quarterly) develops measures of perceived usefulness and perceived ease of use, then examines their relationships with system use. This work distinguishes perceptions that an overall satisfaction questionnaire risks conflating. For an AI assistant, enjoying the experience, considering the tool useful and producing a usable result should therefore be measured separately.
The tempting conclusion
The tempting conclusion would be that, because volunteers want to keep the assistant, the department is ready to adopt it. Yet those volunteers may have more suitable tasks, greater technical familiarity or more freedom to experiment. These are hypotheses to test, not established characteristics of early users. (our executive and employee training programmes)
What these studies cannot establish
These articles concern neither generative assistants nor their no-code development; they provide no quantified effect that can be transferred to a project in Berne. A local pilot may also conflate the tool's effect with that of the additional support given to participants. Comparing volunteers with other users remains descriptive if tasks, trial conditions and available support differ.
A practical check in Berne
Within SHR's programme “Créer ses assistants et agents IA sans coder”, one exercise is to prepare a pilot involving volunteers and a random sample of the other intended users, with consent to participate and refusals documented. In Berne, cases could involve a federal administrative request, public health information, a telecom support ticket or a precision-industry non-conformity report, using fictional or authorised cases and the same support for each group. For each task category, compare the proportion of outputs accepted without major rework against a predefined rubric applied without knowledge of group membership, recording satisfaction separately. A difference between groups would not prove its cause, but would provide a verifiable reason not to generalise results from volunteers alone. To go further: explore the Building AI Assistants and Agents Without Coding training in Bern in the canton of Bern, or browse our executive and employee training programmes in Switzerland.
In pictures: Building AI Assistants and Agents Without Coding in Bern
- building ai assistants and agents without coding
- Bern
- research
- training bern