FREE DM Review Site Registration!
Sign-up today and access DM Review on the Web!

Your FREE registration entitles you to:
FREE email newsletters

FREE access to all DM Review content

FREE access to web seminars, resource portals, our white paper library and more!

   

Predictive Analytics - Addressing the Business Vicissitudes

The Power of Metrics

Several key factors contribute to the success of a predictive analytics implementation. Paramount among those factors is a comprehensive understanding of the company's business goals and business challenges. At the start of a predictive analytics implementation, there is always a tendency to rush forward and select a statistical technique or algorithm before developing a complete understanding of the company's business needs. This "rush to analytical nirvana" often obfuscates the real business challenge at hand. Before we break open the shrinkwrap and attempt to meander through the confusion of neural nets, decision trees and clustering, let us first determine the real business questions that need to be answered:

  • Who are your most profitable customers?
  • How do you make existing customers more profitable?
  • Which customers are at risk for attrition?
  • How do you improve customer service?
  • What is the profile of my existing customers?

These questions represent only a small sample of the type of customer-related challenges that face most companies. Hence, there exists a need to categorize the business challenges within a consistent framework that can then provide a roadmap for selection of the appropriate analytical techniques. As a first step, we will focus on the key high-level business activities:

Classification: The goal of classification analysis is to assign customers to previously defined groups by identifying the attributes which characterize that customer group. The model is developed from historical data by examining already-classified customers and inductively finding a predictive pattern. The identified model and its attributes are then used to classify the new customers. Classification deals with discrete outcomes. Examples of business applications include fraud detection, credit scoring, customer churn and retention.

Clustering: The goal of cluster analysis is to find groups that are very different from each other (heterogeneous), but whose group members are very similar to each other (homogeneous). This is different from classification because the cluster groupings and the attributes of data used for clustering are not previously defined. Typically a business user is needed to interpret the clusters and overlay business relevancy. Examples of business applications include market segmentation, product cross/up-sell and credit policy.

Association: The goal of association analysis is to find trends across a large number of transactions that can be used to understand and exploit customer purchasing behavior. This approach utilizes association rules, which are developed by examining historical data to determine what affinity relationships exist. The best example is analysis of supermarket purchases where a shopping basket is analyzed to develop purchasing behavior rules. There are also situations where purchase activities are related over time, known as sequence analysis. Examples of business applications include market basket analysis, product up/cross-sell, customer portfolio analysis and marketing/promotion campaigns.

Description: The goal of descriptive analysis is to understand the customer data by using a variety of summary and visualization techniques. Summary measures such as averages, means, min/max and standard deviation provide profiles of the distribution of the data. Visualization tools such as frequency histograms, box-and-whisker plots, pareto charts and scatter plots significantly enhance the interpretation of the data. Examples of business applications include customer profiling, share of wallet, money laundering and portfolio management.

Estimation: The goal of estimation analysis is to forecast future outcomes by developing relationships between a customer outcome variable and one or more causal variables. Classification and estimation are often used together to develop models on subsets of the data. Estimation deals with continuously valued outcomes. Examples of business applications include: customer lifetime value, risk profiling and site transversal patterns.

In my next column, we will map the high-level activities to the portfolio of analytical approaches. 


Kent Bauer is the managing director, Performance Management Practice at GRT Corporation in Stamford, CT. He has more than 20 years of experience in managing and developing CRM, database marketing, data mining and data warehousing solutions for the financial, information services, healthcare and CPG industries. Bauer has an MBA in Statistics and an APC in Finance from the Stern Graduate School of Business, New York University. A published author and industry speaker, his recent articles and workshops have focused on KPI development, BI visioning and predictive analytics. Please contact Bauer at kent.bauer@grtcorp.com.

For more information on related topics, visit the following channels:



Industry Vendors