Informally cases can be thought of as groupings of adpositions that produce the same word form. So instead of remembering a bunch of 1:1 mappings from all the adpositions to endings, you first do a 1:N mapping to cases, and then 1:1 from cases to endings.