Challenge configuration
Interactive churn field
Customer decisions
Analysis workspace
| Strategy | Customers contacted | Cost | Expected retained revenue | Net value |
|---|
Detected issues
| ID | Tenure | Charge | Activity | Complaints | Contract | Quality flag |
|---|
Churn probability
A probability estimates how likely a customer is to leave. The threshold converts that score into a stay or churn label.
Precision and recall
Precision limits wasted offers. Recall limits missed churners. Business costs determine which metric deserves greater weight.
Class imbalance
Churners may be rare. Accuracy can look strong while the model misses most customers who actually leave.
False positives
A false positive predicts churn for a customer who stays. It can waste retention budget and reduce campaign efficiency.
False negatives
A false negative misses a real churner. This can lose recurring revenue and valuable future customer relationships.
Feature importance
Feature contributions explain why risk moved upward or downward. They support investigation but do not prove causation.
Data leakage
Leakage exposes information unavailable at prediction time. It makes evaluation unrealistically strong and harms deployment performance.
Customer lifetime value
Lifetime value estimates future economic contribution. High-risk, high-value customers can justify more expensive interventions.
Model bias
Models inherit patterns from data and design choices. Compare subgroup outcomes before relying on automated retention decisions.
Achievements
| Round | Scenario | Mode | Score | Accuracy | Recall | Revenue protected |
|---|