Case: Evaluate a 20% Discount Campaign

Context

Marketing proposes a 20% discount to boost purchases. You are asked to build the business case, design the experiment, and outline rollout and monitoring. Assume a two-sided marketplace with variable contribution margins and payment processing fees.

Tasks

1. Business case and decision rule
   • Should we offer the discount? Formulate an incremental profit framework and a payback-period rule.
   • Explicitly include: coupon cost (20% subsidy), cannibalization (orders that would have happened anyway), and processing/handling fees.
2. Experiment design for causal impact
   • Address self-selection and code leakage via randomized code eligibility, user- or geo-level randomization, and holdout cells.
   • Define primary metrics: incremental gross profit per user and 90-day LTV uplift. Add guardrails: refunds, fraud/abuse, and customer support contacts.
3. Targeting strategy
   • Two-stage approach: first estimate heterogeneous treatment effects (HTE), then deploy targeted eligibility.
   • Compare simple rules (cohorts, RFM) vs uplift modeling; specify how to validate targeting without bias (e.g., nested experiments or interleaved randomized eligibility).
4. Sample size, duration, and data quality
   • Outline sample size and duration requirements. Note data needs: exposure logging, redemption attribution, and handling multiple redemptions.
   • Provide a pre-registered analysis plan, including CUPED variance reduction and cluster-robust standard errors if randomized by geo.
5. Rollout, monitoring, and thresholds
   • Define rollout phases, monitoring cadence, and kill-switch thresholds.
   • Provide an example breakeven calculation for a segment (given AOV, margin, and expected lift) that either justifies or rejects the discount.