Of the insights AI surfaced and acted upon in the last 30 days, 89% were verified accurate once real outcomes came in — 247 production models contributed to the 18.4M predictions scored today. One category, churn risk flags, underperformed its calibration target, with predicted confidence running well ahead of realized accuracy. Business impact: $412K in avoided losses attributed to correctly verified insights this month. Recommendation: recalibrate the Churn Prediction v3.2 confidence threshold before its next verification cycle.

Churn Prediction v3.2 assigns high confidence scores (avg. 88%) that only hold up 76.2% of the time once verified against actual cancellations and renewals — the widest calibration gap of any tracked category. AI attributes this to two factors: the training window pre-dates a Q2 change in the Enterprise renewal contract terms, and 41% of false-positive churn flags involved accounts that renewed after a manual save-play outreach the model has no visibility into. Recommendation: retrain on the last 90 days of renewal data and feed save-play outcomes back into the feature set.

Insights awaiting human confirmation of their real-world outcome

Verification Audit Trail