Explore how time-to-event modeling predicts when customers churn. Master key ideas like hazard, survival curves, and censoring.
In customer churn modeling, ‘censoring’ most often refers to ______.
customers whose churn time is unknown beyond the observation window
customers who churn multiple times in one window
customers removed due to outlier spend
customers excluded for data privacy consent
The hazard function best describes ______.
the cumulative probability of survival until infinity
the probability a new customer signs up at time t
the instantaneous risk of churn at time t given survival until t
the average churn rate over the entire year
A Kaplan–Meier curve steps down when ______.
new customers are acquired
a censored observation appears
a price increase is launched
a churn event occurs in the data at that time
Why can Cox proportional hazards be attractive for churn?
it ignores time-varying covariates by design
it only works with balanced panel data
it requires no assumptions about censoring
it models covariate effects without specifying the baseline hazard
A proportional hazards violation suggests you should ______.
drop censoring records to restore balance
use time-varying effects or stratification to relax constant ratios
remove all covariates that are categorical
switch to linear regression on tenure
Competing risks are relevant when ______.
customers can churn and return in the same day
there is no censoring in the dataset
multiple exclusive churn events can occur and one precludes the others
purchase frequency is very high
For subscription churn, time origin is commonly set to ______.
the quarter when marketing spend peaked
the first time the user installed the app regardless of signup
the start of the customer’s paid tenure or first billing date
the most recent customer support interaction
Left truncation (delayed entry) means ______.
customers who churned before tracking are coded as censored
only left-handed customers are included
the baseline hazard is fixed at zero initially
customers enter the risk set after time zero because earlier history is unobserved
A/B tests for retention can be analyzed with survival methods because ______.
they convert hazards into ROAS by default
they assume no covariates exist
they compare time-to-event outcomes and naturally handle censoring
they require no randomization
Predicting ‘time to churn’ rather than binary churn often helps because ______.
it guarantees proportional hazards hold
it removes the need to monitor cohorts
it eliminates the need for any covariates
it captures when churn is likely, improving timing of interventions
Starter
Great start—review the definitions and practice on small cohorts.
Solid
Nice retention instincts—pressure‑test proportional hazards and competing risks.
Expert!
Excellent—translate hazards into timely lifecycle actions at scale.
