Show TOC

 Customer Data Models for Analysis in Retail

Use

A customer data model is a list of characteristics and key figures about a customer. A customer data model is useful if you want to compare the behavior of different customers in an analysis. In the simplest case, the analysis can be a report or it can be a statistical procedure or a data mining procedure.

If you consider the information that you have about your customers, it consists partly of data that is totally time-independent and partly of data that is loosely time-dependent and can be disregarded in the analysis (master data). You also have data from the complex interaction with the customer that generally consists of sales information and contact information (for example, contacts that you established with the customer during campaigns and contacts that were initiated by the customer). In order to compare two different customers or to compare different periods for one customer, it is necessary to compare the two complex time series. This comparison can be simplified by first defining a common standard, which is the same for all customers in the analysis, and then carrying out the comparison or the analysis on the basis of the common standard. In other words, you define a data model and then use it to compare or analyze your customers.

In SAP BW a data model is represented as a selection of customer attributes and the definition of (restricted/calculated) key figures.

A data model is defined by a query in SAP BW. When modeling customer behavior, the query is used to define and describe how the model is supplied with data.

You need a customer data model if you want to:

  • Investigate or predict the dependency of a customer characteristic on other factors (classification analysis),

  • Divide your customers into different homogenous groups (clustering of customers into customer groups),

  • Predict a key figure for your customers or make a forecast on the basis of historical data (scoring).

For the classification, you require the characteristic for which the dependency analysis is to be carried out for each customer in the analysis, and the characteristics and key figures that you think influence the dependent characteristic. The dependent characteristic is not required for the forecast. In clustering, you need all the characteristics and key figures for each customer that you think influence the formation of customer groups. The process for scoring is the same as for the classification, except in this case you carry out a dependency analysis for a key figure.

In SAP CRM you can use data models in the Segment Builder to define target groups for campaigns.

The choice of a suitable data model is crucial to the quality of an analysis. In this respect it is not possible to make any general statements here for retail. However, a few general comments can be made on the procedure:

  • Customer master data can only be used reliably in the analysis if you are informed of any changes to the data. Even for common demographic data such as address or family status this is not often the case in retail. SAP only recommends you use customer master data if your operative processes are such that you are kept informed of changes.

  • As key figures in retail, you can choose sales for each customer that you restrict to certain time frames and product groups. Hierarchies at article and merchandise category level are useful for this. As an example, in grocery retailing you can use key figures to determine whether a customer has children or whether they have a preference for ecological perishable goods (by looking at the sales for the corresponding merchandise categories). Sales of promotional goods are often used (to identify groups of bargain hunters). If you have divided your products into high- and low-value articles, you can include corresponding sales shares for each customer in your analysis.

  • If you include total sales and sales shares in the analysis, you can include the customer’s sales and their relative share of the total sales (that is, sales share divided by the total sales) in the data model instead. From a mathematical point of view, both these methods are equal (if total sales do not equal 0 - as can result if returns are made), however, the models can behave differently in the analysis. The model with the relative sales shares is generally the preferred method for analysis.

  • The quality of the analysis depends on the suitability of the data model. This can be seen, for example, if a forecast model provides “good” forecasts on test data (that is data for which you carry out the forecast, but for which the result of the forecast is known). This cannot be expressed in absolute figures but you can establish the value of different data models for your analyses. We recommend that for complex dependency analyses, you create several data models and compare the results with one another. You should view making improvements to existing data models as a continuous process. You should also note that established data models can be devalued if the behavior of your customers in your models changes (due to your own measures or due to competitor’s measures).

Prerequisites

Information about sales to your customers at customer/article level (or customer/merchandise category) is available in SAP BW.

Procedure

You define a customer data model in the form of a query in SAP BW.