In Beyond retrieval: A proposal to expand the design space of classification, Melanie Feinberg argues that classifications are not just about efficient retrieval, but about mapping a conceptual space as an active part of problem-solving or design.
A classification highlights connections and contrasts, and fuzzy boundaries, so seems to me to be an obvious tool to help analysis. Comparing different classifications can also illustrate different aspects of an idea or domain. I am very used to the principle of building classifications, seeing how things fit, looking at the things that don’t fit, throwing the classification away, and starting again. You always learn a lot about the topic you are working with in the process.
There seem to be a lot of classifications in Human-Computer Interaction that are used as checklists (things like the DECIDE framework), rather than retrieval tools. It strikes me that library classifications are the special case, rather than “checklist classifications”, which are very common. But then that is just a question of how you classify classifications.