One common mistake customer strategists fall into when using predictive analytics is believing the analytics process alone will find the answer to the key business question. It is easy to think that predictive analytics is all about complex algorithms and data streams, but too often we overlook or underestimate the amount of human interaction that is needed to make a project successful.
A predictive analytics project generally has three cyclical phases: the assessment phase, the analysis phase, and the test and learn phase.
The assessment phase is used to explore the business question. During this phase analysts ask questions such as, what are the key customer interactions that influence the desired outcome? And, for those interactions that are most important, what are some of the decisions, policies, or processes that influence that interaction.
The analysis phase is focused on information gathering, primarily from multiple data sources, and using advanced analytic software. During this phase data scientists are building predictive models and working with the business to operationalize prescriptive actions. This stage includes setting up systems and processes to optimize the prediction.
The test and learn phase involves learning through the changes that have been made and validating the business question. It includes tuning the prediction through additional data sources, insight and feedback from subject-matter experts, and intelligent experiments involving multiple treatments and controls to see which actions yield optimum outcomes. The role of experimentation within the prescriptive process is critical, but is often overlooked.
The payoff of predictive analytics can be huge, but we can't overlook the amount of human involvement that is needed to fully optimize its benefits.