Including the Salesperson Effect in Purchasing Behavior Models Using PROC GLIMMIX

Nowadays, an increasing number of information technology tools are implemented in order to support decision making about marketing strategies and improve customer relationship management (CRM). Consequently, an improvement in CRM can be obtained by enhancing the databases on which these information technology tools are based. This study shows that a salesperson’s personal attitudinal and behavioral characteristics can have an important impact on his sales performance. This salesperson effect can be easily included by means of a generalized linear mixed model using PROC GLIMMIX. This can significantly improve the predictive performance of a purchasing behavior model of a home vending company. INTRODUCTION In an increasingly competitive business environment, a successful company must provide customized services in order to gain a competitive advantage. As a result, many firms have implemented information technology tools to customize marketing strategies in order to build up a long-term relationship with their clients. This study will try to improve such customer relationship management (CRM) models by taking the salesperson effect into account. Traditional CRM models are typically based on variables related to the individual such as socio-demographics, lifestyle variables and the individual past purchasing behavior of the customer. This study suggests that the purchasing behavior of a particular customer can also depend on social surroundings that have an influence during the purchase occasion. In a home vending environment the most important social surrounding is the interaction between the customer and the salesperson. A salesperson’s personal attitudinal and behavioral characteristics have an important impact on his sales performance. because a home vending company decides in advance which salesperson will visit which customer at what time. This makes it possible to already include this knowledge in a highly dynamic model that scores the customers on a daily basis. Hence, PROC GLIMMIX in the SAS® 9.2 program is introduced to capture this effect. This procedure makes it possible to estimate a generalized linear mixed model (i.e. a multilevel model) with a binomial outcome variable. This study will investigate whether data augmentation with the salesperson effect will result in better purchasing behavior prediction. These predictions generated daily can be used for several applications. For example, when the demand is too high to visit every client, these predictions can help to select the most profitable ones. On the other hand, in a situation of overcapacity the salesperson has extra time left, in this situation the predicted probabilities can be used to generate revisit suggestions of the most profitable clients that were not home during the first visit.