View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by Repositório Aberto da Universidade do Porto Analytical Customer Relationship Management in Retailing Supported by Data Mining Techniques Vera L´uciaMigu´eisOliveira A Thesis submitted to Faculdade de Engenharia da Universidade do Porto for the doctoral degree in Industrial Engineering and Management Supervisors Professor Ana Maria Cunha Ribeiro dos Santos Ponces Camanho Professor Jo~aoBernardo de Sena Esteves Falc~aoe Cunha 2012 Abstract Customer relationship management (CRM) has never been as relevant for organizations as it is nowadays. The competitive environment in which com- panies operate is forcing companies to adopt customer centered strategies. In addition, the technologic devolvement observed in recent years enabled companies to keep databases with customer related data. This allows the use of data mining techniques to extract knowledge from these databases in order to gain competitive advantage and remain at the leading edge. This thesis develops a methodology to support CRM in the retail sector, by applying data mining techniques. The research aims to contribute to the improvement of the relationship between retail companies and their cus- tomers. In order to ensure that the methodology proposed can be used in real situations, a company is used as case study. CRM is mainly supported by 4 dimensions: customer identification, cus- tomer attraction, customer development and customer retention. The meth- odology proposed aims to tackle customer identification by segmenting cus- tomers using clustering data mining techniques. This involved an analysis considering two alternative criteria: purchasing behavior and lifestyles. Cus- tomer attraction and customer development are addressed by means of the design of differentiated marketing actions, based on association data mining techniques. Customer retention is approached by the development of models that determine the promptness of customers to leave the company for the competition. These models apply classification data mining techniques. The methodology proposed in this thesis contributes to the marketing liter- ature by exploring several analytical CRM dimensions in the retail sector. Moreover, it provides guidance for companies on the use of analytical CRM to support customers' knowledge achievement and, consequently, enables the reinforcement of the relationship with customers. This thesis also demon- strates the potential of data mining techniques applied to large databases in the context of CRM. i ii Resumo O Customer relationship management (CRM) nunca foi t~aocrucial para as empresas como ´eatualmente. O ambiente competitivo em que as empresas operam tem imposto a ado¸c~aode estrat´egiascentradas no cliente. Para al´emdito, o desenvolvimento tecnol´ogicoverificado nos ´ultimosanos tem permitido `asempresas manter bases de dados com informa¸c~oesrelativas aos clientes. Isto permite o uso de t´ecnicasde data mining para extrair conhecimento dessas bases de dados de forma a obter vantagens competitivas e a ocupar posi¸c~oesde lideran¸ca. Esta tese desenvolve uma metodologia de suporte ao CRM no sector de retalho, usando t´ecnicasde data mining. A investiga¸c~aovisa contribuir para a melhoria da rela¸c~aoentre as grandes empresas de retalho e os seus clientes. De forma a assegurar que os m´etodos propostos possam ser aplicados em situa¸c~oesreais usa-se uma empresa como caso de estudo. O CRM baseia-se essencialmente em 4 dimens~oes:a identifica¸c~ao,a atra¸c~ao, o desenvolvimento e a reten¸c~aodos clientes. A metodologia proposta tem como objetivo abordar a identifica¸c~aodos clientes recorrendo a t´ecnicasde clustering. Isto envolveu uma an´aliseconsiderando dois crit´eriosalterna- tivos: o comportamento de compra e o estilo de vida. A atra¸c~aoe o de- senvolvimento dos clientes s~aoabordados atrav´esdo desenho de a¸c~oesde marketing diferenciadas, baseadas em t´ecnicasde associa¸c~ao. A reten¸c~ao dos clientes ´eabordada atrav´esdo desenvolvimento de modelos de identi- fica¸c~aodos clientes que poder~aovir a abandonar a empresa. Estes modelos baseiam-se em t´ecnicasde classifica¸c~ao. A metodologia proposta nesta tese contribui para a literatura de marketing atrav´esda an´alisede diferentes dimens~oesdo CRM anal´ıticono sector do retalho. Para al´emdito, a metodologia serve como guia `asempresas de como o CRM anal´ıticopode suportar a extra¸c~aode conhecimento e como conse- quentemente pode contribuir para a melhoria da sua rela¸c~aocom os clientes. O trabalho desenvolvido tamb´emdemonstra o potencial das t´ecnicasde data mining aplicadas a grandes bases de dados no contexto do CRM. iii iv Acknowledgments My first thanks goes to Prof. Ana Camanho, whose supervision was crucial for my doctoral work. Thanks for her time and for sharing her knowledge. I also thank Prof. Jo~aoFalc~aoe Cunha for all comments on my work. I thank Prof. Dirk Van den Poel for welcoming me in the UGent and for providing me very good working conditions. I also thank my UGent colleagues for their hospitality, help and friendship. I would like to express my gratitude to the company used as case study and its collaborators, particularly to Dr. Nuno and to Dr. Ana Paula, for providing the data and all necessary information. Thanks for the economic support. It enabled the discussion of the research work by the domain experts. I also thank the Portuguese Foundation for Science and Technology (FCT) for my grant (SFRH/BD/60970/2009) cofinanced by POPH - QREN \Tipolo- gia" 4.1 Advanced Formation, and co-financed by Fundo Social Europeu and MCTES. I thank my colleges and friends from the Industrial Engineering department for all moments of fun over the last years, particularly: Andreia Zanella, Isabel Horta, Jo~aoMourinho, Lu´ısCerto, Marta Rocha, Paulo Morais, Pedro Amorim and Rui Gomes. Marta, I will miss you a lot! I also thank the Professors I have worked with, namely Prof. Ana Camanho, Prof. Jos´e Fernando Oliveira, Prof. Maria Ant´oniaCarravilla and Prof. Pina Marques, for all support concerning all teaching tasks. I also thank Prof. Sarsfield Cabral, head of the department, for his encouragement over these years. I acknowledge D. Soledade, Isabel and M´onicafor all nice chats and friendship. Finally I thank all my family for the support, specially my parents, my brother and my grandmother. I thank all my friends for all moments of fun, which provided me energy to conduct this project. I also thank Andr´efor his company, help, comprehension and love. v Acronyms ANN Artificial Neural Networks AUC Area Under Curve CART Classification And Regrets Techniques CLV Clustering around Latent Variables CRM Customer Relationship Management DIY Do It Yourself DVD Digital Video Device EM Expectation Maximization ERP Enterprise Resource Planning EU European Union GDP Gross Domestic Product ID3 Iterative Dichotomiser3 IT Information Technology KDD Knowledge Discovery in Databases MBA Market Basket Analysis PCC Percentage Correctly Classified POS Point Of Sale RFM Recency Frequency and Monetary RM Relationship Marketing ROC Receiver Operating Characteristic SOM Self-Organizing Map VIF Variance Inflation Factor VLMC Variable Length Markov Chains WOM Word-Of-Mouth vi Table of Contents Abstract i Resumo iii Acknowledgements v Table of Contents vii List of Figures xi List of Tables xiii 1 Introduction 1 1.1 General context . 1 1.2 Research motivation and general objective . 3 1.3 Specific objectives . 4 1.4 Thesis outline . 6 2 Customer Relationship Management 9 2.1 Introduction . 9 2.2 Marketing concept . 9 2.3 Customer relationship management concept . 12 2.3.1 CRM benefits . 13 2.3.2 CRM components . 14 2.4 Analytical CRM . 16 vii TABLE OF CONTENTS 2.4.1 Dimensions . 16 2.4.2 Applications . 19 2.5 Summary and conclusions . 22 3 Introduction to data mining techniques 25 3.1 Introduction . 25 3.2 Knowledge discovery . 26 3.3 Data mining and analytical CRM . 29 3.4 Clustering . 34 3.4.1 Partitioning methods . 36 3.4.2 Model-based methods . 38 3.4.3 Hierarchical methods . 45 3.4.4 Variable clustering methods . 49 3.5 Classification . 49 3.5.1 Logistic regression . 50 3.5.2 Decision trees . 51 3.5.3 Random forests . 58 3.5.4 Neural networks . 59 3.6 Association . 63 3.6.1 Apriori algorithm . 65 3.6.2 Frequent-pattern growth . 66 3.7 Conclusion . 68 4 Case study: description of the retail company 71 4.1 Introduction . 71 4.2 Company's description . 71 4.3 Loyalty program . 73 4.4 Customers characterization . 75 4.5 Conclusion . 79 viii TABLE OF CONTENTS 5 Behavioral market segmentation to support differentiated promotions design 81 5.1 Introduction . 81 5.2 Review of segmentation and MBA context . 82 5.3 Methodology . 85 5.3.1 Segmentation . 85 5.3.2 Products association . 86 5.4 Behavioral segments and differentiated promotions . 87 5.5 Conclusion . 92 6 Lifestyle market segmentation 93 6.1 Introduction . 93 6.2 Methodology . 94 6.3 Lifestyle segments . 95 6.4 Marketing actions . 103 6.5 Conclusion . 105 7 Partial customer churn prediction using products' first pur- chase sequence 107 7.1 Introduction . 107 7.2 Customers retention . 108 7.3 Churn prediction modeling . 109 7.4 Methodology . 112 7.4.1 Partial churning . 113 7.4.2 Explanatory variables . 114 7.4.3 Evaluation criteria . 118 7.5 Partial churn prediction model and retention actions . 119 7.6 Conclusion . 124 ix TABLE OF CONTENTS 8 Partial customer churn prediction using variable length prod- ucts' first purchase sequences 125 8.1 Introduction . 125 8.2 Variable length sequences . 126 8.3 Methodology . 126 8.3.1 Evaluation criteria . 127 8.3.2 Explanatory variables . 128 8.4 Partial churn prediction model . 129 8.5 Conclusion . 132 9 Conclusions 133 9.1 Introduction . 133 9.2 Summary and conclusions .