Essentially you take a bunch of variables (sometimes also over time) and then ask "can I somehow cluster observations according to the variables." Step 2: ??? Step 3: Profit, or at least a publication.