Design metrics and experiment for donation feature

Uber Eats is thinking about introducing a new feature: during checkout, a customer can optionally add a donation to the merchant or to a cause that the merchant has chosen. You are the data scientist responsible for evaluating this feature. Produce a concise plan that an analytics/DS team could execute.

1. Goal and hypotheses. What is the core product objective of this feature? State a primary hypothesis and 2-4 secondary hypotheses, and make sure at least one plausible negative effect is included.
2. Metrics selection. Define your primary (success) metric(s), diagnostic metrics (for understanding mechanisms), and guardrail metrics (for ensuring no harm). Be precise about definitions: unit of analysis, numerator/denominator, time window, and the funnel stage at which each metric is measured.
3. Experiment design. Describe an experiment plan, including: experiment/randomization unit (user/order/merchant), eligibility and experiment population, treatment arm(s) and control, duration and power/MDE considerations, and how you would handle repeated orders, interference/spillover, novelty effects, and heterogeneous treatment effects (e.g., by merchant type or user frequency).
4. Risks and confounders. List the major risks, biases, and marketplace effects (e.g., selection bias, cannibalization, instrumentation bugs, equilibrium effects) that could mislead conclusions, and how you would address each one.
5. Decision framework. Spell out the launch/no-launch criteria, and explain how you would weigh tradeoffs (e.g., conversion down but retention up).

Overview: A PayPal data science onsite case that asks you to design metrics and a randomized experiment for an optional checkout donation feature on a food-ordering platform. Covers goal and hypothesis framing, a primary/diagnostic/guardrail metric hierarchy with precise definitions, user-level experiment design with power and variance reduction, confounder/marketplace-bias handling, and a launch decision framework.