You run product at a test-preparation company. The team tracks daily active users and hours studied, both up strongly. Students pay Rs 18,000 for a nine-month course. Completion of the full syllabus is 31%. Students who complete score, on average, 24 percentile points higher than those who do not. Refund requests run at 8%. The exam is once a year, so success is observed long after the sale.
Propose a north-star metric, its input tree, and guardrails. Say what you would do if the metrics disagreed.
100 points, 60% to pass.
Hours studied is the classic trap: it rewards a product that wastes a student's time. The outcome the student buys is a score, which is observed once a year and far too late to steer by. The design problem is finding a leading indicator that genuinely predicts it — syllabus completion looks right, and the 24-point gap is the evidence, though a strong answer notices that completion may be a proxy for motivation rather than a cause. Inputs: weekly active study sessions, topic mastery, mock-test progression. Guardrails: refund rate and mock scores, so completion cannot be gamed by making the syllabus easier to finish.