Churn Prediction Engine reference profile.
Which customers are likely to churn, and when?
Overview
Ranks the active customer base by churn probability and expected timing so limited CS capacity is spent on the accounts most likely to leave.
Who owns this decision
Chief Customer Officer, VP Customer Success
What this Specialist Model actually does
Decision: Which customers are likely to churn, and when?
The final taxonomy records 6 supporting or external components, but it doesn't promote an Oumi-owned specialist model for this engine.
Who uses this Specialist Model
- Chief Customer Officer
- VP Customer Success
- VP Revenue Operations
- Head of Retention
Where the specialist model sits
The final taxonomy records 6 supporting or external components, but it doesn't promote an Oumi-owned specialist model for this engine.
Feature pipeline from product usage, billing, support, and CRM systems
Supporting component
How Oumi connects to your stack →
Calibration layer per segment
Supporting component
How Oumi connects to your stack →
Continuous retraining on confirmed outcomes
Supporting component
How Oumi connects to your stack →
Dashboard and alerting layer
Supporting component
Dashboard and alerting layer for CS teams
How Oumi connects to your stack →
Gradient-boosted or fine-tuned small-model classifier
Supporting component
Gradient-boosted or fine-tuned small-model classifier for probability scoring
Lightweight LLM component
Supporting component
Lightweight LLM component for support-ticket sentiment and text-signal extraction
How to evaluate and improve it
Evaluation
- AUC-ROC / precision-recall at each risk tier
- Calibration error (predicted probability vs. observed churn rate)
- Lead time accuracy (predicted vs. actual churn window)
Feedback signals
- Actual churn/renewal outcome versus predicted probability
- CSM overrides of the risk tier
- False-positive flags where a "high risk" account renewed
- False-negative flags where a "low risk" account churned unexpectedly
- Model calibration drift by segment over time


