In Customer Intelligence we investigate customer preferences and customer propensities with the goal of offering more targeted products and retain a happier clientele. Since we investigate customers, the input data needs to be customer descriptive. While it is easy to see how demographic features - such as age or gender - can describe our customers, it is more complex to see how a list of contracts can become a customer descriptive feature.
Many years ago, while working with Phil Winters, I have learned that there are three types of descriptive feature areas: revenues, demographics, and behavior.
Demographics features usually are used as they are: no transformation required. Features, such as occupation, salary, gender, and age, come straight from your CRM system and are used as they are.
Revenues related features though do need some flattening, but are still quite easy to calculate and to interpret. In this case, we usually start from a series of contracts on different dates by the same customers. From that list we want to extract only one number, describing how much revenue that customer has brought to the company on a given time frame.
The more complex flattening is usually about behavior, since behavior depends on the domain and involves many aspects of the business. For example, from the list of contracts above, we can derive a measure of loyalty based on the first and last contract date. The customer journey on the company web site, with its number of views and clicks and time spent on each page, represents also a behavioral measure for a web based business. The amount of used electricity or telco service inside specific time windows is another measure of behavior in using the product. And so on. Behavioral features will need designing, domain knowledge, and creativity but will be invaluable in predicting results.