Data mining and optimization : applications in customer portfolio management in the credit card industry
MetadataShow full item record
This dissertation proposes a framework for optimizing customer portfolios. The goal of this framework is to optimize the relationship between a business and its customers, and especially so in the credit card industry. In this framework customers are treated as assets or equity, and this customer equity is measured as Lifetime Customer Value. This dissertation decomposes customer relationships into components, each of which is modeled. These components are translated into cash flows, and the NPV of these cash flows is considered to be the value of the customer as an asset. This dissertation considers customers to comprise a portfolio of assets, and the optimal relationship translates to maximizing the value of the customer portfolio subject to business constraints. The framework laid out in this dissertation can be divided into three parts. The first is the definition of a structure to evaluate the objective of this optimization problem, i.e. the structure to estimate the LCV. The second includes the analytical and predictive modeling methods that feed into the structure of LCV estimation. The third is the optimization formulation that uses the previous two components to prescribe the “optimal business-customer relationship”. Additionally this dissertation takes into account the uncertainty inherent in modeling customer behavior and incorporates this into the optimization formulation. This dissertation combines datamining and optimization into a decision making framework by incorporating results from supervised and semi-supervised learning techniques (used in modeling and clustering) in an optimization formulation. Neural networks are used to model key aspects of consumer behavior, and the errors in these models provide a basis for deriving data scenarios for a stochastic programming formulation of the problem. The modeling and optimization framework is illustrated throughout by a real example using data from a project at CompuCredit, aimed at designing customer incentives to activate their credit cards.