Health Model Governance

Health model governance defines the rules that maintain model quality over time — who can modify signal weights, how often the model is recalibrated, what quality thresholds trigger model review, and how model changes are tested before deployment. Without governance, the health model degrades as market conditions change, new signals emerge, and outdated weights produce increasingly inaccurate scores. With governance, the model evolves systematically and its predictive accuracy improves over time.

1. Establish a Model Owner One person owns the health model — responsible for its accuracy, its calibration, and its evolution. The model owner reviews predictive accuracy quarterly, proposes weight adjustments, tests changes against historical data, and deploys updates. Without a named owner, the model becomes an orphaned system that nobody maintains and everybody distrusts.
2. Define Calibration Cadence The model is recalibrated quarterly by comparing predicted outcomes against actual outcomes. If the model predicted 15 accounts at high churn risk and 12 actually churned, the model is well-calibrated. If it predicted 15 at high risk and 3 actually churned, the model is over-predicting risk and the weights need adjustment. The calibration cadence ensures the model stays accurate as the business evolves.
3. Require Testing Before Deployment Weight changes are tested against historical data before they are applied to production scores. The test answers: "If these weights had been in effect for the past year, would the model have been more or less accurate?" Only changes that improve historical accuracy are deployed. This prevents well-intentioned adjustments from degrading model quality.