Primer: Predictive AnalyticsBy Kevin Fogarty | Posted 2004-12-01 Email Print
Re-Thinking HR: What Every CIO Needs to Know About Tomorrow's Workforce
Learn about the next step after data mining.
What is it? Sophisticated algorithms designed to sift through a data warehouse and identify patterns of behavior that suggest what offers customers might respond to in the future, or which customers you may be in danger of losing. For example, when sifting through a bank's data warehouse, an application using predictive analytic techniques might recognize that customers who cancel an automatic bill payment or automatic deposit often move to another bank within a certain period of time.
Isn't that data mining? It's related, but with a more advanced degree in statistics. Data mining relies on a database manager who can structure a specific question and define the fields in the database from which the answer will come. For example: "Of the accounts that were closed last month, how many included an automatic deposit and were located in Chicago?"
A predictive analytic query would include parameters, but fewer specifics. It might, for example, analyze patterns of activity in all accounts to identify geographic locations or changes in accounts. Then, it would correlate those factors to find patterns related to accounts that were closed, or that were expanded. That approach would not only identify that a cancelled automatic payment was a danger sign, but that location was not a relevant factor.
What business problem would it solve? A report from Forrester Research's Giga Information Group says predictive analytics can identify potential incidences of fraud by highlighting changes in existing or new accounts that are consistent with previous offenses; reduce inventory and shipping costs by more accurately predicting sales volumes and locations; and increase the profitability of each customer by recognizing opportunities to sell add-on products or services, or to keep them from defecting.
How do I buy it, and for how much? Predictive analytics is a technique, not a product. So you'd buy it from companies that sell business-intelligence software, marketing-analysis applications, data-mining tools and other related products. Unica Corp. of Waltham, Mass., for example, sells predictive modeling tools within its "enterprise marketing management suite" of applications, which are designed to help identify market opportunities, valuable customers, trends in customer behavior and tactics for addressing those patterns. Business Objects of San Jose, SPSS of Chicago, San Francisco-based KXEN, and other customer relationship management (CRM) and data-mining software companies offer products with predictive analytic capabilities. They're also available within databases from IBM, Oracle and Teradata.
Costs vary with the customer's size and software type. CRM or database costs depend on the size of the implementation and what kind of applications you build on top of them. Unicawhose products come in discrete chunks, making it easy to identify and draw cost comparisons among the predictive analytic modulesestimates a company with revenue in the tens of millions would pay about $50,000 for the software, while one with $1 billion in revenue would pay in the hundreds of thousands.
What's the downside? To make accurate predictions, your data has to be pretty clean. That means going through your data warehouse to make sure John Smith, J. Smith and John. Q. Smith are the same person, with the same buying history. You may need to scrub the data you have, or retrain the people who enter the data so they understand the importance of data format rules. You may also discover you need more information than you have, meaning you'll have to buy demographic data from external service companies such as credit bureaus.