production-mlops · Deployment Readiness

Churn Prediction v2 — Production Readiness

Model: churn-xgb-v2 · Skill: deployment-readiness-for-ml · Command: /mlops-ops · Generated 2026-06-22
Deployment Status
READY
Shadow mode complete (7 days). Canary at 10% approved. High-LTV recall gate failing — deploy with segment-level monitoring and automated rollback trigger at recall < 0.68.

Serving Configuration

Serving modeBatch daily
Inference cadence02:00 UTC, nightly
Artifact formatONNX (churn-xgb-v2.onnx)
InfraAWS Batch + S3 output
Prediction volume~42k customers/run
Output sinkRedshift predictions table

Monitoring SLOs

MonitorThresholdAlert
Input feature drift (PSI)< 0.20Active
Prediction distribution shiftKL < 0.15Active
Batch job success100%Active
High-LTV segment recall≥ 0.68 (alert) / ≥ 0.75 (gate)Watchlist
Null rate in key features< 2%Active

Rollout Strategy

1
Shadow Mode — Completed
Run model in parallel with v1 for 7 days. Compared prediction distributions, validated no systematic bias vs v1 baseline. No rollback triggered.
Done · 2026-06-15
2
Canary — 10% traffic · In progress
Route 10% of customer scoring to v2. Monitor recall on High-LTV segment daily. Automated rollback if recall drops below 0.68 for 2 consecutive runs.
Active since 2026-06-20
3
Promote to 100% — Gated on recall gate
Full promotion requires High-LTV recall ≥ 0.75 sustained for 5 days at canary. Current value: 0.71. Estimated gate pass: 1–2 sprints with feature engineering improvement (planned for Sprint 15).
Blocked on recall gate

Rollback Plan