The product wants to "predict churn". How do you define the label?
Answer it out loud before you open anything. The value of the flags below is in comparing them to what you actually said — including whether you asked about the data before naming a model.
A streaming subscription service has monthly plans, annual plans, free trials, involuntary churn from failed payments, and customers who pause for a season and come back. A director has asked for "a churn model". The data science team has three different definitions in three notebooks, and the numbers they report are not comparable.
React to this
Say what you would question, what you would trust, and what you would need to know first.
Three notebooks, three labels (illustrative): nb_A: churn = status == 'cancelled' at any point in the next 90 days AUC 0.91 nb_B: churn = no active subscription 30 days after next renewal date AUC 0.74 nb_C: churn = (cancel_reason IS NOT NULL) OR (payment_failed_3x) AUC 0.97 Features common to all three: plan_type, tenure_days, watch_hours_30d, support_tickets_30d, cancel_reason, last_payment_status, pause_flag.
What it is really testing
Whether the candidate treats the label as a design decision with consequences rather than a column that exists. The strong answer works from the decision backwards, fixes a prediction time and a horizon, handles the edge cases explicitly, and knows that a label that is easy to compute is often a label that is useless to act on.