Dries F. Benoit

Benefits of quantile regression for the analysis of customer lifetime value in a contractual setting: An application in financial services

The move towards a customer-centred approach to marketing, coupled with the increasing availability of customer transaction data, has led to an interest in understanding and estimating customer lifetime value (CLV). Several authors point out that, when evaluating customer profitability, profitable customers are rare compared to the unprofitable ones. In spite of this, most authors fail to recognize the implications of these skewed distributions on the performance of models they use. In this study, we propose analyzing CLV by means of quantile regression. In a financial services application, we show that this technique provides management more in-depth insights into the effects of the covariates that are missed with linear regression. Moreover, we show that in the common situation where interest is in a top-customer segment, quantile regression outperforms linear regression. The method also has the ability of constructing prediction intervals. Combining the CLV point estimate with the prediction intervals leads to a new segmentation scheme that is the first to account for uncertainty in the predictions. This segmentation is ideally suited for managing the portfolio of customers.

Bayesian adaptive Lasso quantile regression

Recently, variable selection by penalized likelihood has attracted much research interest. In this paper, we propose adaptive Lasso quantile regression (BALQR) from a Bayesian perspective. The method extends the Bayesian Lasso quantile regression by allowing different penalization parameters for different regression coefficients. Inverse gamma prior distributions are placed on the penalty parameters. We treat the hyperparameters of the inverse gamma prior as unknowns and estimate them along with the other parameters. A Gibbs sampler is developed to simulate the parameters from the posterior distributions. Through simulation studies and analysis of a prostate cancer dataset, we compare the performance of the BALQR method proposed with six existing Bayesian and non-Bayesian methods. The simulation studies and the prostate cancer data analysis indicate that the BALQR method performs well in comparison to the other approaches.

Improving customer retention in financial services using kinship network information

This study investigates the advantage of social network mining in a customer retention context. A company that is able to identify likely churners in an early stage can take appropriate steps to prevent these potential churners from actually churning and subsequently increase profit. Academics and practitioners are constantly trying to optimize their predictive-analytics models by searching for better predictors. The aim of this study is to investigate if, in addition to the conventional sets of variables (socio-demographics, purchase history, etc.), kinship network based variables improve the predictive power of customer retention models. Results show that the predictive power of the churn model can indeed be improved by adding the social network (SNA-) based variables. Including network structure measures (i.e. degree, betweenness centrality and density) increase predictive accuracy, but contextual network based variables turn out to have the highest impact on discriminating churners from non-churners. For the majority of the latter type of network variables, the importance in the model is even higher than the individual level counterpart variable.

Enhanced decision support in credit scoring using Bayesian binary quantile regression

Fierce competition as well as the recent financial crisis in financial and banking industries made credit scoring gain importance. An accurate estimation of credit risk helps organizations to decide whether or not to grant credit to potential customers. Many classification methods have been suggested to handle this problem in the literature. This paper proposes a model for evaluating credit risk based on binary quantile regression, using Bayesian estimation. This paper points out the distinct advantages of the latter approach: that is (i) the method provides accurate predictions of which customers may default in the future, (ii) the approach provides detailed insight into the effects of the explanatory variables on the probability of default, and (iii) the methodology is ideally suited to build a segmentation scheme of the customers in terms of risk of default and the corresponding uncertainty about the prediction. An often studied dataset from a German bank is used to show the applicability of the method proposed. The results demonstrate that the methodology can be an important tool for credit companies that want to take the credit risk of their customer fully into account.

A Bayesian approach for incorporating expert opinions into decision support systems

Interest in the use of (big) company data and data-mining models to guide decisions exploded in recent years. In many domains there are human experts whose knowledge is essential in building, interpreting and applying these models. However, the impact of integrating expert opinions into the decision-making process has not been sufficiently investigated. This research gap deserves attention because the triangulation of information sources is critical for the success of analytical projects. This paper contributes to the decision-making literature by (a) detailing the natural advantages of the Bayesian framework for fusing multiple information sources into one decision support system (DSS), (b) confirming the necessity for adjusted methods in this data-explosion era, and (c) opening the path to future applications of Bayesian DSSs in other organizational research contexts. In concrete, we propose a Bayesian decision support framework that formally fuses subjective human expert opinions with more objective organizational information. We empirically test the proposed Bayesian fusion approach in the context of a customer-satisfaction prediction study and show how it improves the prediction performance of the human experts and a data-mining model ignoring expert information. This paper introduces a decision support framework to fuse information sources.Fusing big data with human opinions ensures higher-quality decisions.The paper demonstrates the advantage of the Bayesian machinery for information fusion.

Identifying influencers in a social network

Individuals influence each other through social interactions and marketers aim to leverage this interpersonal influence to attract new customers. It still remains a challenge to identify those customers in a social network that have the most influence on their social connections. A common approach to the influence maximization problem is to simulate influence cascades through the network based on the existence of links in the network using diffusion models. Our study contributes to the literature by evaluating these principles using real-life referral behaviour data. A new ranking metric, called Referral Rank, is introduced that builds on the game theoretic concept of the Shapley value for assigning each individual in the network a value that reflects the likelihood of referring new customers. We also explore whether these methods can be further improved by looking beyond the one-hop neighbourhood of the influencers. Experiments on a large telecommunication data set and referral data set demonstrate that using traditional simulation based methods to identify influencers in a social network can lead to suboptimal decisions as the results overestimate actual referral cascades. We also find that looking at the influence of the two-hop neighbours of the customers improves the influence spread and product adoption. Our findings suggest that companies can take two actions to improve their decision support system for identifying influential customers: (1) improve the data by incorporating data that reflects the actual referral behaviour of the customers or (2) extend the method by looking at the influence of the connections in the two-hop neighbourhood of the customers. A Shapley value method for finding top influencers in a customer network is proposed.Using referral behaviour data improves the selection of top influencers.Simulation-based influence spread overestimates actual influence spread.Considering the influence of two-hop neighbours improves the influencer selection.

Integrating expert knowledge and multilingual web crawling data in a lead qualification system

Qualifying prospects as leads to contact is a complex exercise. Sales representatives often do not have the time or resources to rationally select the best leads to call. As a result, they rely on gut feeling and arbitrary rules to qualify leads. Model-based decision support systems make this process less subjective. Standard input for such an automated lead qualification system is commercial data. Commercial data, however, tends to be expensive and of ambiguous quality due to missing information. This study proposes web crawling data in combination with expert knowledge as an alternative. Web crawling data is freely available and of higher quality as it is generated by companies themselves. Potential customers use websites as a main information source, so companies benefit from correct and complete websites. Expert knowledge, on the other hand, augments web crawling data by inserting specific information. Web data consists of text that is converted to numbers using text mining techniques that make an abstraction of the text. A field experiment was conducted to test how a decision support system based on web crawling data and expert knowledge compares to a basic decision support system within an international energy retailer. Results verify the added value of the proposed approach. A decision support system for lead qualification is developed.It uses web crawling data augmented with expert knowledge as input.Integrating expert knowledge increases the quality of the decision support system.A real-life test validates these results.

Improving direct mail targeting through customer response modeling

Data-mining algorithms (CHAID, CART and neural networks) perform best.Logistic regression and linear discriminant analysis are statistical alternatives.Quadratic discriminant analysis, naive Bayes, C4.5 and k-NN algorithm perform badly. Direct marketing is an important tool in the promotion mix of companies, amongst which direct mailing is crucial. One approach to improve direct mail targeting is response modeling, i.e. a predictive modeling approach that assigns future response probabilities to customers based on their history with the company. The contributions to the response modeling literature are three-fold. First, we introduce well-known statistical and data-mining classification techniques (logistic regression, linear and quadratic discriminant analysis, naive Bayes, neural networks, decision trees, including CHAID, CART and C4.5, and the k-NN algorithm) to the direct marketing community. Second, we run a predictive benchmarking study using the above classifiers on four real-life direct marketing datasets. The 10-fold cross-validated area under the receiver operating characteristics curve is used as evaluation metric. Third, we give managerial insights that facilitate the classifier choice based on the trade-off between interpretability and predictive performance of the classifier. The findings of the benchmark study show that data-mining algorithms (CHAID, CART and neural networks) perform well on this test bed, followed by simplistic statistical classifiers like logistic regression and linear discriminant analysis. It is shown that quadratic discriminant analysis, naive Bayes, C4.5 and the k-NN algorithm yield poor performance.

Incorporating sequential information in bankruptcy prediction with predictors based on Markov for discrimination

In this paper we make a contribution to the body literature that incorporates a dynamic view on bankruptcy into bankruptcy prediction modelling In addition to using financial ratios measured over multiple time periods, we introduce variables based on the Markov for discrimination (MFD) model. MFD variables are able to extract the sequential information from time-series of financial ratios and concentrate it in one score. Our results obtained from multiple samples of Belgian bankruptcy data show that using data collected from multiple time periods outperforms snap-shot data that contains financial ratios measured at one point in time. In addition, we demonstrate that inclusion of MFD variables in non-ensemble bankruptcy prediction models considered in the study can lead to better classification performance. The latter type of models, despite not achieving the top performance based on metric considered in our study, can still be used by practitioners who prefer simpler, more interpretable models. Bankruptcy should be viewed as a dynamic process.Markov for discrimination (MFD) allows for dynamism at the level of input variables.MFD variables capture information in time series of financial ratios.MFD variables lead to better performance of bankruptcy prediction algorithms.

Preventing Customers from Running Away! Exploring Generalized Additive Models for Customer Churn Prediction

Nowadays, companies invest in a well-considered Customer Relationship Management strategy. One of the cornerstones of CRM is customer churn prediction, where one tries to predict whether or not a customer will leave the company. This study focuses on how better to support marketing decision makers in identifying risky customers by using Generalized Additive Models (GAM). Compared with logistic regression, a GAM relaxes the linearity constraint that allows for complex nonlinear fits to the data. The contributions to the literature are threefold: (i) it is shown that a GAM is able to improve marketing decision making by better identifying risky customers; (ii) it is shown that a GAM increases the interpretability of the churn model by visualizing the non-linear relationships with customer churn, identifying a quasi-exponential, a U, an inverted U, or a complex trend, and (iii) marketing managers are able to increase business value significantly by applying a GAM in this churn prediction context.