Tracking of consumer behaviour in e-commerce

With the increasing amount of transactional data available on e-commerce websites, it is possible to gain a deeper insight into the dynamics of a given business. Based on the profiles of the different types of customer and the changes they undergo over time, new marketing strategies can be developed to target specific groups of users. The aim in this work is to estimate the future state of a customer and decide whether to target that customer or not. In the first part of the paper, we describe our proposed algorithm which uses hierarchical clustering and a hidden Markov model (HMM). The clustering can have one (non-augmented) or two levels (augmented). We compare the augmented and non-augmented method to a benchmark with synthetic and real data to show that our model outperforms the others in predicting future customer behaviour. In the second part of the paper, we use a decision-theory tool to estimate whether it is financially beneficial for the business to adopt our proposed model, as opposed to a less complex one. We conclude that there might not be any benefit at all, even though the model is more accurate in the predictions. This will depend on the utility functions at stake.