Predicative analytics sifts though current and historical data to predict future events or behaviors. It incorporates statistical and data mining techniques to determine patterns, highlight risks and identify opportunities.
It doesn’t just extrapolate future performance based on past performance, although that is an input. It also identifies the relationships across many factors to lay out not only what is likely to happen, but also why it may occur along with potentially how to alter the outcomes.
Where there’s data warehousing, there’s predictive analytics. It is increasingly in many industries including retail, telecom, pharmaceutical, insurance and financial services. These industries typically leverage predictive analytics to analyze and predict consumer and business behaviors along with economic predictions.
The most successful and rewarding (business ROI) predictive analytical processes include: enhancing customer retention and loyalty; identifying cross-sell and up-sell opportunities; identifying customer lifetime value; fraud detection; underwriting; portfolio analysis; and direct marketing.
Predictive analytics has made the transition from an emerging technology into the mainstream. It has also moved from long-term strategic predictions to much more immediate and tactical or operational analysis. Companies in insurance (claims), financial services (credit cards) and health care (claims) are incorporating operational predictive analytics in such areas as fraud detection. These reduce losses that cost both these companies and their customers.
The bedrock of predictive analytics is data and lots of it. These processes consume large amounts of historical and current data. In addition to past events (facts in the dimensional modeling lingo), many other factors related to customers, products, other businesses, and the economy (dimensions) are also needed to identify relationships and formulate predictions. The companies with the most success in getting business returns from predictive analytics have a significant commitment in data integration that feeds these analytical processes.
Predictive analytics, however, is only as good as its data. Comprehensive, consistent, clean and current data enables the statistical and data mining techniques being used in predicative analytics.
2 Comments
Rick, excellent posting! I agree that data quality is one of the foundations for predictive analytics, but I would add open standards for models and integration as another key driver for further adoption.
For predictive analytics, the DMG (http://www.dmg.org) has defined the Predictive Modeling Markup Language (PMML), a vendor-independent standard which allows the exchange of predictive models among various data mining solutions. It is supported by a growing number of vendors and creates an ecosystem in which the user has a true choice between solutions for data mining, modeling, deployment, and scoring.
Once you standardize your decision models in PMML format, you can deploy your models in minutes using a scoring engine or an enterprise decision management platform. Please see Buying Predictive Analytics Like Books which describes how we at Zementis leverage cloud computing to lower the barrier of entry for predictive analytics.
Businesses increasingly recognize the value of operational predictive analytics. With a framework that includes a high-performance data warehouse, open standards to exchange predictive models among applications, and cloud computing as a cost-effective way to integrate predictive analytics in real-time, adoption will gain momentum.
Dr. Zeller, Thanks for your kind words.
I agree with your statement that “open standards for models and integration as another key driver for further adoption of predictive analytics.” Open standards should expand the use of predictive modeling by making these models more accessible, more robust (by enabling more people to augment the open models) and hopefully to reduce the costs of predictive modeling.
There is certainly shareable and reusable models and heuristics across business processes and industries. For too long predicitive analytics and data mining have been dominated by highly customized solutions. This has resulted in higher costs and longer development times to achieve success.
Data quality and open models should expand and ensure the success of predictive modeling.
Rick Sherman