Most health scores were built once, weighted by gut feel, and left to drift. By the time an account turns red, the customer has often already decided to leave. A growing number of CS teams are replacing these static, rules-based scores with predictive models that update continuously from product usage, support activity, sentiment in calls and emails, stakeholder changes and commercial data. The results can be striking, but so are the failure modes: models nobody trusts, scores that cannot explain themselves, and CSMs who quietly go back to their spreadsheets. This panel brings together CS leaders who have rebuilt their health scoring with AI to talk candidly about what worked and what didn’t. They’ll discuss which signals actually predict churn and expansion, how they got data from product, support and RevOps into one place, and how they made a constantly changing score something CSMs and executives act on rather than argue with. Key takeaways: - Which real-world signals carry the most predictive weight, and which popular inputs turned out to be noise - How to move from a static weighted score to a model that learns, without losing the team’s trust in it - Why explainability matters as much as accuracy, and how to show CSMs the reason behind every score change - How health scores are evolving from a churn warning into a guide for expansion, renewal forecasting and next best action