You are Camila Estévez-Pardo, 34, Head of Growth Analytics at Canasta, a Bogotá online-grocery subscription. Your data-science team just delivered a churn-prediction model — and it works. The CMO wants to point the entire retention budget at it on Monday. You have the weekend to decide how.
The situation
Canasta delivers weekly grocery boxes to about 64,000 subscribers across Bogotá, Medellín and Cali. The business is growing, but monthly churn of 6.5% is the single biggest drag on it — every month roughly 4,000 subscribers cancel, and acquiring their replacements costs far more than keeping them would.
The board approved a retention budget: COP 168M a month — enough for about 6,000 vouchers worth COP 28,000 each (roughly a fifth of a box), sent to subscribers the company believes are about to leave. Your data-science team built a gradient-boosted churn model. It scores every subscriber 0–100 on the probability of cancelling next month, and it is genuinely good: 0.87 AUC, validated on held-out data. The CMO's plan is simple — "each month, take the 6,000 highest-scored subscribers, send them the voucher. We finally have a model. Let's use it."