AI assistants: which data do they actually need to collect?
Building an assistant without coding makes it easier to design a tool, but also a new point of information collection. The management question is specific: which data should be requested to deliver the intended service without making additional disclosure the default?
What the literature establishes
Tamara Dinev and Paul Hart (2006, Information Systems Research) examine a model in which willingness to provide personal information for electronic transactions depends, among other factors, on perceived risks, trust and expected benefits. Their findings support an interpretation based on trade-offs rather than a simple distinction between users who care about privacy and those who do not. For assistant designers, this suggests a precaution: agreeing to provide information does not demonstrate that the information is necessary.
A decision shaped by context
Alessandro Acquisti, Laura Brandimarte and George Loewenstein (2015, Science) review research showing that disclosure decisions are characterised by uncertainty and dependence on context. How a request is presented can influence behaviour without changing the objective usefulness of the data requested. A welcoming form or a smooth conversation is therefore not, on its own, an indicator of informed choice.
The premature conclusion
One might conclude that an assistant should always request less information. The useful conclusion is more demanding: distinguish data needed for the task from data the interface merely encourages users to provide. A client's name may be unnecessary when drafting a generic answer to a procedural question; when handling an individual case in an authorised environment, some identifying data may be essential. (our executive and employee training programmes)
What these studies cannot settle
These publications do not compare no-code generative assistants used in organisations, so they cannot quantify how reducing input fields affects their quality. Decisions made by employees subject to instructions and a hierarchy are not equivalent to those made by consumers either. Finally, removing a visible identifier guarantees neither the anonymity of a narrative nor control over access and retention.
A practical check in Zurich
As part of SHR's programme “Créer ses assistants et agents IA sans coder”, a relevant exercise for financial services, insurance, tech and the European headquarters of international groups in Zurich is to test an assistant on representative fictional cases, with and then without each category of personal data requested. For each pair, a domain expert assesses the answers against a predefined rubric without knowing which version was used; the team records the proportion of cases in which removing the data preserves an acceptable answer, along with the failures observed. Categories whose removal does not impair results on these cases become candidates for removal from the form, ahead of a separate review of security, compliance and retention requirements. To go further: explore the Building AI Assistants and Agents Without Coding training in Zurich, or browse our executive and employee training programmes in Switzerland.
In pictures: Building AI Assistants and Agents Without Coding in Zurich
- building ai assistants and agents without coding
- Zurich
- research
- training zurich