Episode Details

What if the AI system your company built to stop people from leaving is actually the thing making them leave? That's the attrition prediction paradox — and it's showing up at HR teams that have invested seriously in people analytics. In this episode, we break down three ways these systems go wrong: the self-fulfilling prophecy, where flagging someone as a flight risk changes how managers treat them until the employee actually quits; Goodhart's Law, where attrition scores become KPIs that get gamed instead of fixed; and class imbalance, where too many false positives burn manager trust in the tool entirely. The good news is that attrition prediction isn't a broken technology — it's a powerful capability that gets misapplied. The difference between teams that extract real value and teams that create more churn comes down to one thing: treating the model's output as the start of a human-led retention conversation, not the end of one. We also cover what best practice actually looks like today — explainability layers, human-in-the-loop decision-making, and interventions that address root causes rather than risk scores.