Segmentation of Known Customers Using
Clustering Methods 
Purpose
You want to divide your customers into homogenous groups, on the basis of predetermined factors.
Unlike a classification on the basis of a predefined key figure as in the ABC classification, here you want to divide customers into different segments automatically, on the basis of a complex customer data model. It is important for the forming of different groups that different combinations of characteristics or key figures and their features are unique. The advantage of the analysis is that you can identify clear customer behavior in a customer group determined automatically using the clustering method. In grocery retailing, for example, a cluster can show that above-average sales are made on children’s products and perishables. You could interpret such a cluster as “family shopping”. The result of a cluster analysis not only provides you with information on which other characteristics the “family shopper” typically demonstrates, but also how large this segment is.
In retail, you generally have greatly varying amounts of data available for each customer. It makes little sense, however, to compare customers about whom you have only very limited information, with your regular customers. More complex data models are more appropriate for analyzing regular customers, who you can determine using an ABC classification together with a loyalty analysis. When using complex data models you should limit the customers that you include in the analysis accordingly.
In order to carry out the analysis, you
need to save the characteristics and key figures that you want to include in
the analysis in a Customer Data
Model. In the Data Mining Workbench you then need to define a new
clustering analysis and use the query in SAP BW that converts your customer
data model. To make a prediction, you have to use clustering that has already
been carried out; this means that the 
training step must
have been successful.
Process Flow
...
1. You define a customer data model for your analysis in SAP BW in the form of a query.
2. In the Data Mining Workbench you perform the clustering analysis (either as training or as a prediction).
3. You return the result of the clustering analysis to SAP BW.
See also: