Advanced Business Analytics

Analytics for business as we know it in terms of predictive analytics probably started in ancient Roman times when the concept of insurance was first created. The Romans were particularly concerned with the rituals around death and dying and therefore first created the concept of funeral insurance. Today, the selling of funeral insurance through direct TV is often viewed by the regulators and the public alike as a deceptive substitute for life insurance targeted to low fixed-income elderly folks.

The chapter proposes a severity of dormancy model (SDM), which allows the potential dormancy and the extent of dormancy to be modeled separately and is attempted for one of the key retail businesses for a large European business. The methodology developed in this chapter challenges the conventional wisdom of building a dormancy model using a logistic regression technique. It recognizes the existence of a group of potential dormant customer who would never go dormant under any circumstances. Most importantly, it not only recognizes this subset of customers, but also explicitly models the probability of the extent of dormancy to depend on customer attributes. Finally, the chapter successfully demonstrates the improvement achieved by the SDM over the conventional technique to capture the severity of dormancy.

Similar to the SDM, this chapter attempts to introduce a class of model known as double hurdle models (originally proposed by Cragg (1971)), which allows the potential attrition and the extent of attrition to be modeled separately. This double hurdle model of attrition is attempted for a European auto finance portfolio. The methodology developed in this chapter identifies a group of potential attritor who would never attrite under any circumstances. Most importantly, it not only recognizes this subset of customer, but also explicitly models the degree of attrition to depend on customer attributes. Identification of this group of customers is crucial to any consumer finance business, since it provides the basis of efficient retention tactics and profitable target population. The chapter successfully demonstrates the improvement achieved by the double hurdle model over the conventional logistic regression technique in all the segments of attrition (over the different degrees of attrition).

Focusing on the consumer finance business of a large financial company of Europe, this chapter analyzes the efficiency of advertising expenditures on three media—TV, print, and the Internet. Using a data envelopment analysis (DEA) method, it assesses the overall efficiency of the media and the effective combinations in attracting new loan applications. It develops an optimal media mix model to evaluate the effect of different media expenditure in getting the approved new loans to the business. Total cost per week for TV, print, and Internet is used as inputs and two outputs; namely, approved new loans and calls received per week are used within a DEA framework to estimate the efficiency score. The incremental benefits from this analysis were estimated at 38 %. The efficiency score developed in this chapter clearly recognizes the best practices weeks, with that we can identify the best combination of media, which serves the purpose best in terms of acquiring new loan applications. Finally, a media mix builder is developed to guide the business for the future strategies. Advertising analytics is gaining momentum as a powerful means to efficient targeting. Sir Martin Sorell, CEO of WPP Group, one of the world’s largest advertising agencies, calls econometrics the holy grail of advertising. Moreover, several top advertising agencies have now created teams of econometricians to do this type of analysis for clients. Therefore, this research and further research in this area are necessary to insure that advertising resources are well spent.

Most retail businesses operate in a non-contractual settings, and this relationship with the customers poses difficulties in differentiating between the customers who have attrited voluntarily and those who are in the middle of their long cycle transaction behavior. Therefore, formulating an effective CRM strategy in retail poses a significant challenge. This chapter proposes a DGP (data generating process)-based predictive strategy with the past purchase transaction data, which would help the business to improve the overall marketing performance with minimum data requirement. In contrast to many existing RFM (recency, frequency, and monetary value)-based models, a set of model with strong underlying behavioral model is proposed, thereby providing a greater insight into the customer decisions. The approach basically predicts a customer’s future purchase money value by combining the three key transaction factors, viz. recency, frequency, and monetary value which is further combined into a more powerful single predicted value (PRFM) for each customer. This represents an original contribution as many retailers are making decisions with RFM, but these are anchored on a static metric based on looking at past behavior and are not predictive in nature. Furthermore, when they do try to render RFM predictive, the methods are often ad hoc, and therefore, they are usually difficult to implement in practice. The final and most important characteristic of this model is the extensibility. Though the models depend only on three key customer attributes, R, F, and M, it can be easily extended to incorporate other customer attributes of interest by running the algorithm for each subsegment. Suggestions for future research include the adaptation of these techniques to all types of general-purpose revolving credit cards which are being issued by most banks and consumer finance companies.

In direct marketing campaign, response models are often developed only based on the data of selected population. Since the propensity to respond depends on selection, this introduces a possibility of bias in the estimates of the response model. This chapter tries to apply a bivariate probit model with partial observability to correct the bias arising due to the sample selection and has applied the proposed model to a non-risk situation, viz. marketing campaign. This represents an original contribution as most marketing quantitative professionals to date have not been concerned with this sample selection bias due to the unstructured nature of applying models in a marketing context and also the fact that risk modeling approaches have had more time to mature from a research standpoint. Apart from addressing the selection bias, the model has helped to identify the customers who are likely to respond if selected for mailing which has led to a significant expansion of the existing mailing universe. This has a direct financial benefit to the home lending firm as it will serve to increase their customer share of wallet and subsequent profits.

Price optimization solutions presented in this chapter provide an analytic approach that helps the business to improve margins and increase volumes. The chapter proposes a comprehensive pricing framework which not only maximizes short-term gains but also addresses critical value enhancing CRM issues, such as cross-sell, up-sell, and better life cycle management through retention. The significant contribution of the chapter involves developing a framework that explicitly and transparently takes into account the price response and adverse elasticity concepts. Also in order to successfully capture the consumer behavior, this chapter introduces advanced modeling techniques such as the double hurdle model, in contrast to the more traditional logistic models, and demonstrates its efficiency to model attrition and risk. Additionally, this chapter introduces a sophisticated clustering technique called “genetic algorithm” for segmentation analysis. Finally, based on the insights from this analysis, a dynamic optimization tool is developed to effectively improve the risk-adjusted profit for the business.

The focus of the chapter is to model the cannibalization effect of a cobranded bank card upgrade program launched on Europe’s retail cards data. Since most of the retail businesses operate in a non-contractual setting, differentiating between the customers who are loyal and those who will continue to maintain their in-store spends even after the upgrade is difficult. The chapter develops an in-store intensity model that challenges the conventional wisdom of building a model using a logistic or OLS regression technique. The methodology developed in this chapter clearly recognizes the existence of a group of potential customers who would direct all their spending on the out-of-store outlets and none on the in-store outlets, by identifying the “flipping point” of their intensities. The intensity model not only recognizes existence of this group of customer but also explicitly models the probability of actual purchase activity to depend on customer attributes, successfully demonstrating a significant improvement over the conventional technique by capturing the extent of in-store intensity and the selection criterion required to pinpoint the most profitable customers in a gainful manner. This methodology will certainly satisfy the retailers in terms of ensuring there is no drop in sales; however, further research is needed for looking beyond just maintaining in-store spend intensities on upgraded customers and optimizing the entire decision process across all of the key drivers of profitability in credit cards.

Missing values and sparse data often challenge the reliability of statistical analysis in terms of biased parameter estimates and degraded confidence intervals, thereby leading to false inferences and suboptimal business decisions. To managers in the consumer data analytics field, the challenge faced by missing and limited data is nothing novel, and many powerful techniques of analysis and data management are available to them. However, the choice of adequate management practices is far from optimal. This chapter proposes an integrated approach by jointly treating the missing data and sparse data problems, using approximate Bayesian bootstrap (ABB) and Bayesian (HB) modeling. Therefore, the chapter addresses these two key challenges and corrects the bias formed, by extrapolating information from the sparse and missing data onto a large sample. The proposed method is illustrated by computation of price elasticity models for a leading consumer finance business on data that suffers from both missing and sparsity issues. The results presented illustrate the superiority of the model in taking better decisions in consumer data analytics. In contrast to the point estimate generated using traditional price elasticity models, the proposed model helps to make a better inference on the price elasticity estimates through a probability density function as it generates a distribution of price elasticity. Further expansion of the principle illustrated here will auger a powerful business optimization possibility and should be a fruitful area of future research.

With an ever-increasing need for firms to analyze data being collected from various sources such as the Internet and other forms of e-commerce, there is a greater need for more improved segmentation techniques for differentiated marketing programs aimed at maximizing revenues and profitability. K-means clustering is a popular technique for segmenting large data sets. Recently, algorithms mimicking the behavior of ant colonies have been shown to bring significant improvements to the K-means clustering algorithm and other methods of knowledge discovery in databases. These techniques were developed by imitating the behavior of real ants for finding the shortest path from their nests to the food source. This chapter represents an application that aims to cluster a data set by means of an ant colony optimization algorithm. It also increases the working performance of this algorithm used for solving the data clustering problem by proposing a multipronged foraging approach, resulting in the globally optimal solution and showing the advantage in the performance due to suggested technique. A limitation of this study is the generalizability of the results to other data sources as this algorithm was only tested in production on financial services data. Further research is necessary on additional sources of data from other domains.

In the modern environment of service-based marketing techniques, maximizing customer lifetime value has evolved into a crucial objective of CRM, in order to obtain profits from creating and sustaining long-term relationships with their customers. This chapter makes a contribution by reviewing the various CLV techniques and modeling advances in this area and in addition highlights the direction for development. It specifically addresses the key challenges in the literature with regard to integrating dynamic, macroeconomic aspects into the CLV which has become imminent given the current economic and financial turmoil.