As organizations continue to move up the “people analytics maturity curve”, the capability and appeal of using advanced statistical modeling techniques to predict who is likely to leave and why, is becoming more prevalent. However, creating statistical models to predict attrition risk is not enough. Two critical issues prevent organizations from harnessing maximum value from these efforts. First, you cannot sacrifice on robustness of your models. This may seem obvious but often times organizations overestimate the validity of their data, and in doing so inevitably compromise the integrity and utility of their predictive models. When adopting predictive analytics, many organizations make the mistake of oversimplifying the process by running bivariate correlations on a handful of HRIS data fields, often focusing on a single data source, or performing simple psychometric assessments. Unless the organization adopts a more “complete” stance by gathering data from different sources, internal or external, and then validate their model predictions against actual results, the credibility and value delivered by these models will remain questionable.