Implementing Neural Modeling in Financial CRM
At the same time, banks increasingly are realizing that the more they know about their customers the more potentially profitable those customers are. The meteoric rise in the development of data warehouses is being fueled by decreases in the cost and availability of computing power and electronic storage, and a marked increase in the banks' ability to analyze the data using a wide range of data mining tools. Selecting the right tools is essential to the banks' ability to offer the right product to the right customer at just the right time.
Data mining with the power of neural modeling. Until recently, marketers have relied on traditional data mining tools such as statistical and rules-based marketing models. These predictive models can be extremely effective at customer segmentation and in modeling situations where there is limited historical data on the individual, such as a marketing campaign targeted at new account acquisition. In more complex marketing campaigns, however, rules that are too tight can mean that only those accounts that match the specific rule will be targeted for the campaign. As a result, many prime prospects for a specific product or service can be overlooked. Conversely, if the rule set is too general, the predictive quality of the information provided by the system is less than optimum. Rules-based systems also can become large and unwieldy as the models are refined and enhanced over time.
Neural network modeling offers clear advantages in marketing programs that hinge on predicting a customer's future purchasing behavior and needs. Widely regarded for their ability to accurately pinpoint fraudulent credit card transactions among the myriad of daily authorization activity, neural models have helped bank card issuers save billions of dollars. But despite the extensive use of neural networks in managing the inherent risks of their businesses, financial service companies are just starting to realize the benefits that neural networks can bring to bear on their revenue-generating marketing programs.
In a marketing application, neural models can be trained on an individual's established track record with the institution to predict customer credit risk, the probability of attrition or retention, cross-sell and up-sell opportunities, overall profitability and other issues that dramatically impact the return on investment of the marketing campaign. Neural software models that combine, for example, customer account information, credit bureau data, and historical data such as transaction types and amounts, types of goods purchased, merchant information, increases in card-usage frequency and cumulative amount of purchases, size and frequency of payments and credit line usage information can be developed to identify new opportunities and influence future buying behavior.
By uncovering even the most complex, multidimensional relationships hidden within customer data, neural models offer an ideal marketing and customer intelligence tool. This combination of predictive neural modeling and burgeoning customer data can provide banks with a major advantage in their quest to build market share and maximize the profitability of their card portfolios.
Neural models also differ from traditional predictive models in their ability to dynamically learn from the data they analyze, allowing them to keep pace with changes in data patterns that a predictive model might miss after it has been in operation for some months.
Like most artificial intelligence systems, neural networks are not designed to replace or work like human beings. Rather, they are designed to perform tasks for which humans are not as well suited. For example, a neural network can make millions of quick decisions about which customers to include in a campaign without tiring or losing attention. While the neural network is making decisions based on the customer information, transactions, payment information and other data, marketers can focus on refining the campaign messages and promotions to ensure that the response rates warrant continued investment in the marketing program or product.
Perhaps the single misconception of data mining with neural networks is that the models are obscure, magical black boxes that do not justify the expense. Today's best neural network data mining tools, however, provide marketers with a prioritized list of factors that contributed to the model's decision. From an ROI perspective, financial services marketing professionals can learn a lot from their peers in risk management, and evaluate neural network modeling based on the return on their investment in the technology, and the incremental lift in the response rates of their one-to-one marketing programs.
Thomas Spillane is director of marketing for financial services at decision support solutions provider Nestor, Providence, RI.