##### Scenario
Predicting whether an enterprise customer will renew their Google Meet contract.
##### Question
Describe how you would build a model to predict customer retention. Which features would you engineer and why? When would logistic regression be sufficient and when would you prefer more complex models (e.g., GBM, NN)? How would you evaluate and compare model performance?
##### Hints
Cover sampling window, target definition, feature importance, calibration, AUC, business lift, interpretability vs performance trade-off.
Quick Answer: Evaluates renewal prediction modeling for enterprise Google Meet contracts. Strong answers define leakage-safe timelines, engineer usage, reliability, support, contract, and customer-health features, compare logistic regression with complex models, and evaluate calibration, lift, and business intervention value.
Build a Model to Predict Customer Contract Renewal
You are designing a model to predict whether an enterprise customer will renew a Google Meet contract. The model should score accounts early enough for customer-success or sales teams to intervene.
Constraints & Assumptions
Treat the prediction unit as an account-contract unless stated otherwise.
Define observation window, blackout period, prediction time, and outcome window.
Avoid leakage from renewal negotiations or post-renewal behavior.
Include business usefulness, calibration, and interpretability.
Clarifying Questions to Ask Guidance
What counts as renewal versus churn, downgrade, partial renewal, or late renewal?
How far before contract end does the business need a prediction?
What interventions are available for accounts at risk?
Are contract, usage, support, pricing, and customer-success data available?
Part 1 - End-to-End Setup
How would you define the target and sampling timeline?
What This Part Should Cover Guidance
Account-contract unit, feature window, blackout period, prediction date, renewal window, and label definition.
Handling grace periods, downgrades, expansions, and renewals after contract end.
Part 2 - Feature Engineering
Which features would you engineer and why?
What This Part Should Cover Guidance
Product usage, meeting reliability, active seats, feature adoption, support tickets, admin activity, account size, industry, contract terms, price changes, sales activity, and customer health.
Trends, seasonality, missingness, and leakage checks.
Part 3 - Model Choice
When is logistic regression sufficient, and when would you prefer more complex models?
What This Part Should Cover Guidance
Logistic regression for interpretability, calibration, small data, and linear effects.
Gradient boosting or other models for nonlinear interactions and larger datasets.
Calibration, explainability, and operational constraints.
Part 4 - Evaluation
How would you evaluate and compare models?
What This Part Should Cover Guidance
AUC, PR-AUC, log loss, calibration, lift at top risk deciles, business lift, feature importance, and segment performance.
Time-based validation and monitoring drift.
Value of interventions and thresholding.
What a Strong Answer Covers Guidance
A strong answer designs a leakage-safe renewal prediction workflow and evaluates models by both statistical performance and whether the scores drive useful retention actions.
Follow-up Questions Guidance
How would you handle accounts with very little usage history?
What if the model is accurate but sales does not trust it?
How would you measure whether interventions based on the model work?