Understanding the Predictive Analytics Lifecycle

Chapter 5Modeling

How analysts don’t just look at the present . . . but peek into the future

The year was 1998. I was just about to start a new analytics job on Paulista Avenue in Sao Paulo, the main drag, a fancy boulevard nestled in between megamalls and soccer stadiums. My first day, I walked in and warmed up the crowd. I made business casual small talk with my new staff. “Good morning. How are you? By the way, one way to identify which model to develop and use is to ask business questions.” I’ve always liked to cut straight to the point. Ultimately advanced analytical systems are designed to create sound statistical models.

These models are designed using standard research methods. The methods help answer critical business questions with extreme value relevance. I rounded up a Brazilian health insurance team around 11 A.M. and jumped right in. “Case management and clinicians will want to identify high-risk (high-cost) patients, identify the best intervention, determine which patients are likely to have a better outcome with an intervention, identify the factors that influence quality of care measures, predict likely diagnosis or next care need, and streamline patient assessments. Which one of these questions should we tackle first?”

Sensing less excitement than I anticipated, I continued, “We can continue identifying potential targets: Underwriting and actuarial staff ...