Predictive modeling uses a variety of analytical techniques to make predictions about the future based on current and historical data. These predictions are expressed as numbers that correspond to the likelihood a particular event, opportunity, or behavior will take place in the future. Predictive modeling can be used in making increasingly effective and individualized decisions about the treatment customers. These models analyze the customers’ past performance in order to assess how likely a customer is to exhibit a specific behavior or respond to a specific offer.
There are several mature predictive modeling applications. One of these is credit scoring. Scoring models estimate the likelihood that that a customer will make future credit payments on time based on their credit history and application information. Another relatively mature predictive modeling application is in targeted marketing which involves using consumers’ past purchasing history and response rates along with demographic, geographic and other relevant characteristics in order to estimate the likelihood that customers will respond to particular marketing offers.
These mature applications represent the tip of the iceberg with respect to the overall opportunity for using predictive modeling for the optimization of customer relationships. While most of these mature approaches have been run off line, leading organizations are beginning to embed predictive models in customer-facing processes in ways that generate revenue opportunities and control risks in real time during live transactions and interactions with customers. Competitive pressures are driving companies to personalize the way they manage customer relationships. This is increasingly possible as:
- Companies have invested heavily in the integration and quality of their customer data.
- More powerful predictive modeling tools are available including advanced statistical regression and time series approaches, as well as, emerging machine learning techniques such as neural networks, radial basis functions, and support vector machines. These machine learning techniques provide powerful tools for automated pattern recognition and prediction.
We see five major application categories of predictive modeling for optimizing customer relationships:
- Valuation. Leading companies are beginning to actively measure and manage the asset value of their customer relationships. The first and most basic question is: what’s the lifetime value of this customer? Based on a customer’s unique characteristics and transaction pattern what types and magnitude of investment is justified? When we make an investment in this customer, does it generate transaction patterns that reflect an increase in their value?
- Customization. Uniquely targeting consumers with the products, services, and experiences they value and types of offers they are likely to respond to can lead to significant revenue growth while reducing acquisition costs. This goes well beyond traditional targeted marketing and cross-selling. Examples: Amazon and NetFlix recommendation engines. Intelligent wardrobing recommendations made by call center agents at Victoria’s Secret Direct. Leading financial services company that predicts a unique “next logical sale” to offer to each customer when they call for any service issue or inquiry.
- Pricing. Many businesses have to account for unique customer risk and price based on the cost of covering that risk. For example, auto insurance providers must accurately determine the amount of premium to charge to cover each automobile and driver. More effective predictive modeling can streamline the process of customer acquisition, by predicting the risks of a particular customer and making more effective pricing decisions.
- Retention. Too many businesses try to retain customers only after the customer attempts to terminate their service. At this stage, changing the customer’s mind can be expensive. In addition, silent attrition, where customers slowly but steadily reduce usage, is a problem faced by a wide range of companies. Leading organizations are adopting more proactive retention strategies by creating early warning systems that detect any significant change is customer behavior that may indicate either a service or retention issue. These companies then take preemptive measures to retain customers and address any latent service issues. These attrition models examine each customer’s transaction history, service usage, and service performance in order to estimate the likelihood that customer will want to terminate service in the near future. Example: We helped JM Family Enterprise develop an Early Warning System that signals changes in customer behavior… regular call reports are generated for each account manager indicating what action they should take with these customers.
- Fraud Detection. Fraud includes inaccurate credit applications, fraudulent transactions, false insurance claims, and identity theft. Fraud undermines the profitability of companies and drives up the costs of goods and services for customers. Property and casualty insurance fraud is approximately $30 billion a year. Health care fraud is approaching $100 billion. Credit card fraud is estimated to cost $1-2 billion a year. Tens of thousands of consumers are victims of identity theft. Increasingly effective predictive models are being used to help quickly identify fraudulent activity without increasing the number of false positives that impact the customer experience.
(Note: I’ve posted a follow on to the above post: Adaptive Customer Profiling: Integrating Qualitative and Quantitative Customer Analytics)