Context
Didi Chuxing's Growth team wants to test a new coupon module on the Didi Rider app homepage. The module highlights a limited-time ride discount before the user enters the ride-request flow, and leadership wants to know whether it drives incremental completed trips without hurting marketplace health.
Hypothesis Seed
The proposed change is a more prominent homepage coupon card with personalized copy and countdown urgency. Product believes this will increase completed-trip conversion among exposed riders by encouraging more users to start and finish a ride request, especially among price-sensitive occasional riders.
Constraints
- Eligible traffic: 1.2 million rider app opens per day in the target cities
- Only 35% of app opens are from riders eligible for coupons
- Maximum experiment duration: 14 days, because the campaign budget and city-ops calendar are fixed
- False positives are costly because over-discounting can reduce contribution margin and create driver supply-demand imbalance
- False negatives are also meaningful because coupon inventory is limited and the team must decide whether to scale before a holiday peak
- The experiment must run in 8 cities with strong weekday/weekend seasonality
Deliverables
- Define the single best primary metric for this experiment, plus 2-4 guardrail metrics and 1-3 secondary metrics. Be explicit about why your primary metric is better than simpler but potentially misleading alternatives such as coupon click-through rate.
- State the null and alternative hypotheses, choose the unit of randomization, and explain whether the test should be one-sided or two-sided.
- Calculate the required sample size using an explicit MDE. Assume baseline completed-trip conversion from eligible homepage opens is 12.0%, and the smallest worthwhile lift is 3% relative.
- Propose a pre-registered analysis plan: statistical test, peeking policy, multiple-comparison policy, and how you will handle city/device heterogeneity.
- Name the main experimentation pitfalls for this marketplace setting, including at least one issue related to interference or SUTVA, and give a clear ship / don't-ship rule that respects guardrails.
