"Computing the Right Pitch"

Computerworld (03/14/05) P. 36; Monash, Curt A.

Analyst Curt Monash describes predictive analytics as "a replacement phrase for 'data mining' [that] roughly equates to 'applications of machine learning and/or statistical analysis to business decisions.'" Business decisions, as defined in most current and short-term applications, are forms of small group marketing, and Monash lists questions that business analytics attempts to address, such as which customers are likely to churn; what types of offers will attract new customers or retain old customers; which prospective customers are most probabilistically profitable, nonprofitable, and churn-threatening; and what content should be shown to particular Web surfers when the next page is served. Monash says the information used to answer such questions can be culled from a diverse array of sources, including transactional data, customer contacts, and third-party data. The difficulty resides in the mathematical methods employed to address predictive questions: The process involves formalizing the problem as one of clustering or classification, and the answer as an algorithm that places each customer or prospect into one of the limited number of buckets. Data on previous prospects and customers serves as evidence used to build the algorithm. Producing such algorithms is beyond the capabilities of conventional statistical techniques, and calls for methods that include neural networks, support-vector machines, and linear algebra. Monash says the best algorithm for any given problem usually consists of a "complex hierarchy of 'elementary' algorithms."

Full Article: http://www.computerworld.com/databasetopics/businessintelligence/
story/0,10801,100322,00.html