Wouter Buckinx

Customer base analysis: partial defection of behaviourally loyal clients in a non-contractual FMCG retail setting

Customer Relationship Management (CRM) enjoys increasing attention as a countermeasure to switching behaviour of customers. Because foregone profits of (partially) defected customers are significant, an increase of the retention rate can be very profitable. In this paper, we focus on the treatment of a company’s most promising customers in a non-contractual setting. We build a model in order to predict partial defection by behaviorally-loyal clients using three classification techniques: Logistic regression, ARD Neural Networks and Random Forests. Classification accuracy (PCC) and area under the receiver operating characteristic curve (AUC) are used to evaluate classifier performance. Using real-life data from an FMCG retailer we show that future partial defection can be successfully predicted. Similar to direct-marketing applications, we find that past behavioral variables, more specifically RFM variables (recency, frequency, monetary value) are the best predictors of partial customer defection.

Predicting online-purchasing behaviour

This empirical study investigates the contribution of different types of predictors to the purchasing behaviour at an online store. We use logit modelling to predict whether or not a purchase is made during the next visit to the website using both forward and backward variable-selection techniques, as well as Furnival and Wilson’s global score search algorithm to find the best subset of predictors. We contribute to the literature by using variables from four different categories in predicting online-purchasing behaviour: (1) general clickstream behaviour at the level of the visit, (2) more detailed clickstream information, (3) customer demographics, and (4) historical purchase behaviour. The results show that predictors from all four categories are retained in the final (best subset) solution indicating that clickstream behaviour is important when determining the tendency to buy. We clearly indicate the contribution in predictive power of variables that were never used before in online purchasing studies. Detailed clickstream variables are the most important ones in classifying customers according to their online purchase behaviour. In doing so, we are able to highlight the advantage of e-commerce retailers of being able to capture an elaborate list of customer information.

Predicting Customer Loyalty Using The Internal Transactional Database

Loyalty and targeting are central topics in Customer Relationship Management. Yet, the information that resides in customer databases only records transactions at a single company, whereby customer loyalty is generally unavailable. In this study, we enrich the customer database with a prediction of a customers behavioral loyalty such that it can be deployed for targeted marketing actions without the necessity to measure the loyalty of every single customer. To this end, we compare multiple linear regression with two state-of-the-art machine learning techniques (random forests and automatic relevance determination neural networks), and we show that (i) a customer’s behavioral loyalty can be predicted to a reasonable degree using the transactional database, (ii) given that overfitting is controlled for by the variable-selection procedure we propose in this study, a multiple linear regression model significantly outperforms the other models, (iii) the proposed variable-selection procedure has a beneficial impact on the reduction of multicollinearity, and (iv) the most important indicator of behavioral loyalty consists of the variety of products previously purchased.