Context

StreamBox built a binary classification model to predict which subscribers are likely to churn in the next 30 days so the retention team can send discounts and outreach. The model is a logistic regression classifier trained on 1.2M historical users, but the marketing lead says the current evaluation is unclear because the team reports only accuracy.

Current Performance

Confusion Matrix (Validation Set: 100,000 users)

The Problem

Leadership sees 91% accuracy and believes the model is strong, but the retention team argues it is missing too many customers who actually churn. You need to explain which metrics matter most here and whether this model is good enough to deploy as-is.

Requirements

1. Interpret the current metrics and explain why accuracy alone is misleading.
2. Identify the key tradeoff between precision and recall for this churn use case.
3. Recommend which evaluation metrics should be prioritized and why.
4. Suggest specific ways to improve model performance or decision thresholding.
5. Explain how you would validate whether changes actually improve business outcomes.

Constraints

• Retention budget allows outreach to at most 10,000 users per month.
• A false positive costs about $4 in discount/contact cost.
• A false negative means losing an average of $65 in monthly gross profit.