Consider a store selling books by mail. The customers are identified, so we don't have that problem.
However, we can suppose that many characteristics of customers are not identified such as age, income, occupation. Literary tastes can be identified in so far as they correspond to a tendency to buy books in predefined classifications. However, the data may allow inferring new classifications with respect to which the customers behave more consistently than with respect to the traditional classifications.
Some classes of customers are leaders in that their preferences today can be used to predict the market for a book later. Identifying such customers and classes of customers may be useful.
The above phenomena--age, etc.--are not in the data per se, but they are rather close to it. Their identification should not be as ambitious a project as identifying the customers of a supermarket. Almost all of the computations will involve the individual customers. The leadership phenomenon involves more but still has a rather simple character.