Predicting Women’s Department Purchases in a Retail Store by using the SEMMA Methodology

As one of the main retail companies in Chile, we are currently facing interesting challenges. Nowadays customers have so many choices than ever to select what, where, how much and how to purchase. For this reason, it is crucial to leverage what technology has to offer in terms of analytics in order to improve value of average tickets and avoid inactivity of customers. For example, it is possible to know with a high degree of certainty which products offer to customers based on buying patterns to improve the cross-selling process. It is made through propensity predictive models, a set of mathematical and statistical techniques, that allows to know what customers can buy in function of historical data. One of our focus to improve the business is the Women Department because it is currently our most powerful department in terms of transactions originated by customers. It raises the need to implement an analytical model focused on this department for establishing what offer is the most appropriate for our customers according to buying patterns of customers, augmenting the likelihood that a customer comes back to the stores. Those patterns are calculated based on demographic, transactional data and any other interaction that our customers have had with our stores. The predictive model we used is the logistic regression and it was executed following the SEMMA methodology considered by SAS ® for projects in SAS Enterprise Miner ®. This work is presented according to the SEMMA methodology and contains the following sections. The description of data to use or samples are expressed in the Sample section. The second section is titled Exploration. It contains all the details of descriptive analysis carried out to formulate the predictive model and to select the input variables. Variable transformations needed for good model implementations are described in the third section called Modification. The Modeling section establishes some relevant aspects and preliminary results of model estimation and steps to improve that model. Finally, the validation section shows results of the final model, measured in a validation sample. In this section a number of criteria to evaluate the model and the final solution are also detailed. Customer Intelligence SAS Global Forum 2013