648 Prague Economic Papers, 2019, 28(6), 648–669, https://doi.org/10.18267/j.pep.714 PREDICTIVE PERFORMANCE OF CUSTOMER LIFETIME VALUE MODELS IN E-COMMERCE AND THE USE OF NON-FINANCIAL DATA Pavel Jasek, 1 Lenka Vrana, Lucie Sperkova, Zdenek Smutny, Marek Kobulsky * Abstract The article contributes to the knowledge of customer lifetime value (CLV) models, where extensive empirical analyses on large datasets from online stores are missing. Based on this knowledge, practitioners can decide about the deployment of a particular model in their business and academics can design or enhance CLV models. The article presents predictive performance of selected CLV models: the extended Pareto/NBD model, the Markov chain model, the vector autoregressive model and the status quo model. Six large datasets of medium and large-sized online stores in the Czech Republic and Slovakia are used for a comparison of the predictive performance of the models. Online stores have annual revenues in the order of tens of millions of euros and more than one million customers. The comparison of CLV models is based on selected evaluation metrics. The results of some of the models which use additional non-financial data on customer behaviour – the Markov chain model and the vector autoregressive model – do not justify the effort which is needed to collect such data. The advantages and disadvantages of the selected CLV models are discussed in the context of their deployment. Keywords: CLV models, forecasting, online marketing management, e-commerce, methodology, online shopping, online marketing. JEL Classification: C53, C55, M21, M31 1. Introduction Models of the customer lifetime value (CLV) primarily allow companies to better allocate resources and formulate strategies in marketing (Ferrentino et al., 2016). Companies use CLV mainly in the area of customer segmentation to build long-term relationships with customers and effectively manage investments in marketing activities (Weng and Huang, 2018). However, use of CLV can help answer a number of very different questions such * Pavel Jasek, Faculty of Informatics and Statistics, University of Economics, Prague, Czech Republic (pavel.jasek@vse.cz); Lenka Vrana, Faculty of Informatics and Statistics, University of Economics, Prague, Czech Republic (lenka.vrana@vse.cz); Lucie Sperkova, Faculty of Informatics and Statistics, University of Economics, Prague, Czech Republic (lucie.sperkova@vse.cz); Zdenek Smutny, Faculty of Informatics and Statistics, University of Economics, Prague, Czech Republic (zdenek.smutny@vse.cz); Marek Kobulsky, Faculty of Nuclear Sciences and Physical Engineering, Czech Technical University in Prague, Prague, Czech Republic (MerrekSK@gmail.com). This article was prepared thanks to the grant of the University of Economics, Prague, under Grant agreement number F4/18/2014.