Make the successful transition to actionable predictions

Does your organization need to use predictive analytics? What lessons have you learned?

 Homeowner insurance claims in 2011 were a record-setting $105 billion, according to Marsh's 2012 Insurance Market Report (PDF). This made it challenging for insurance companies to manage their policy risk, and the risk exposure was further amplified by the fact that homeowners insurance (when compared to auto) was historically more volatile in its claims performance.

The business challenge created an opportunity for big data analytics that insurers could apply to their homeowner insurance portfolios in order to better assess risk - and to predict where claims are most likely to occur.

Predictive analysis

As a result, insurance companies now use scoring methodologies to predict future loss-ratio performance and also the likely behaviors of those whom they insure. This governs how insurers determine individual homeowner premiums, deductibles, and discounts on policies. Insurers also use predictive data modeling to determine the likelihood of severe claims, fraud and even litigation risks. The move to predictive analytics extends to the evaluation of weather patterns, which are now changing with global warming.

Meanwhile in the education sector, the University of Telecommunications in Leipzig, Germany, wanted a predictive means of determining what future hiring and talent needs in the telecommunications industry were likely to be. The goals were to be proactive in curriculum revisions, to stay in step with the job market, and to offer students the right types of training for the job market.

For its analytics, the university processed thousands of online job postings, analyzing unstructured job market data in the process. It aggregated employment requirements across the entire telecommunications industry so it could monitor emerging trends that might not readily be obvious from simply perusing employment listings.

One of the first predictive discoveries was that the demand was decreasing for degrees in electrical and communications engineering, and instead shifting toward informatics, business administration, and business informatics. This intelligence enabled the university to change its curriculum at a record pace (the average new course was being launched in 2.5 months instead of in 12 months - a 76% improvement). The improved agility in curriculum adjustments was noticed by students, also. The university saw a 300 percent increase in demand for the new courses.

Define roles clearly because big data success requires it

Business value but there is a flip side

Both the insurance and the education use cases demonstrate the importance and the immediate business value that can be derived from effectively harnessing big data into predictive analytics that are actionable and that prepare organizations for better future outcomes.

On the flip side though, life is not always predictable. Predictive analytics would not likely have selected Beethoven, who was deaf, as a potential composer. Nor would it have predicted that Bill Mazeroski, a scrappy second baseman who only hit 138 homers in his 16-year major league baseball career, would be the one to win the 1960 World Series against New York for the Pittsburgh Pirates with a home run.

John Elder, CEO at Elder Research, a data mining firm, said that, "The vast majority of analytic projects are riddled with mistakes." Elder estimates that 90 percent of the projects he sees are "technical successes," but that only 65 percent of that 90 percent are ever implemented by their organizations.

Good news

Much of the problem rests in cultivating clean and representative data that can be fed into analytical processes to guarantee accurate results. In other cases, predictive analytics projects are scoped so large that it becomes difficult to control the analytics process in order to get to the outcomes that were initially expected of them.

The good news is that there are lessons we are already learning about predictive analytics that can be used to improve the odds of predictive analytics success. These include:

  1. Developing well defined business cases for predictive analytics that don't try to do too much;
  2. Taking the time needed to qualify and sanitize the data that will flow into these analytics so that you know upfront that you have the best possible data that you can have; and
  3. Leaving some room for failure and even unexpected results - because the irony of predictive analytics is that it can be unpredictable.

Also read: