DATA MINING, MODELING, & PREDICTIVE SERVICES: EXTENDING FUNCTIONAL DECOMPOSITION
by John K. Thompson

In the previous article I discussed how monolithic data mining systems are being deconstructed and reconstituted as smaller units of functionality that are optimized for execution of portions of the data mining and predictive services tasks. I also alluded to how these smaller units are serving different constituencies. In this article, I extend that idea further.

Broadly speaking, analytical systems have failed to deliver on the promise of providing relevant causal information to people who are required to make decisions on a day to day basis. This is true for a multiplicity of reasons too numerous to elaborate on in this forum, but let's assume this premise as an underlying fact.

In observing highly competent, successful decision makers, it is noted that these individuals possess a talent for examining a wide array of information sources, extracting the salient points, synthesizing the information, and producing highly accurate decisions. This mode of operation is lacking in the current analytical systems.

Typically, analytical systems are implemented in a manner such that the results from each system are isolated from other analytical systems and from the transaction processing systems that maintain the day to day records and operations for the company. This isolation and the difficulty of navigating through far-flung systems tends to slow down the best decision makers and completely thwart those who are less adept or tenacious.

One of the driving forces in decomposing large data mining systems is to enable the embedding of the germane functions into existing transaction and analytical systems. For example, breaking the data mining and predictive services functions apart creates the ability to embed predictive services into transaction processing systems such as credit card authorization systems, retail sales, customer service and others. By embedding the predictive services engine and lightweight predictive models, the analytical information produced is presented to the decision-maker in context of current organizational knowledge.

In conjunction with presenting the new information in context, the ability to leverage these new lightweight analytical systems is spread across the organization. Rather than requiring business analysts to gather, analyze and disseminate their findings, personnel at the customer touch points can be utilizing the systems and information to make faster and more informed decisions on how best to serve that individual customer that is engaged in the current transaction or interaction.

With this type of operation, data mining begins to more closely achieve one of the ultimate goals of technology; that is to deliver truly valuable and useful information while becoming less visible and more ubiquitous. For personnel who are responsible for optimizing customer satisfaction, there should be no concern or knowledge that the new element appearing on their existing output device is generated through data mining. All they need and want to know is that this new element of information helps them help the customer.

In my next article, I will complete the explanation of the extent of functional decomposition and introduce some of the links between the decomposed systems.

---

John Thompson, Vice President - Marketing, Magnify, Inc. You can reach me at jkt@magnify.com