Preference models reveal trade-offs only when the choices resemble real decisions.
Conjoint analysis can be powerful because it asks respondents to make choices rather than rate every feature as important. But its apparent precision can obscure poor research design.
Attributes must reflect trade-offs a buyer actually encounters. If levels are implausible, descriptions ambiguous, or the choice setting artificial, estimated preferences may look stable while describing no real market behavior.
Qualitative exploration and pilot testing should shape the experiment. Analysts should examine whether respondents understood the options and whether modeled choices track observable behavior or credible holdout tasks.
The value of conjoint is not a universal ranking of features. It is a structured way to compare plausible offerings under defined conditions, with uncertainty and limitations plainly stated.