Evaluate marketplace interventions

You are a data scientist at a two-sided delivery marketplace. Answer the following product analytics and experimentation cases. For each case, define the product goal, the causal estimand, the primary and guardrail metrics, the appropriate randomization unit, likely sources of bias or interference, and how you would make a launch decision. 1. **Driver coupon** - The company wants to offer coupons or incentives to drivers to improve marketplace outcomes. - How would you evaluate whether the driver coupon works? - Consider short-term effects on supply and completed trips, as well as long-term effects on driver retention and marketplace profitability. 2. **Bunch delivery** - The company introduces a feature that allows one driver to deliver multiple orders on the same route. - How would you evaluate the impact of bunch delivery? - Consider tradeoffs among driver utilization, delivery time, customer experience, cancellation rate, and contribution margin. 3. **Delivery fee vs. service fee** - Why might a marketplace charge a **delivery fee** and a **service fee** separately instead of using a single all-in fee? - If the company wants to change the delivery fee, how would you analyze the likely impact on conversion, basket size, driver supply, and unit economics? 4. **Lower ETA computation time** - The ETA model is changed so that prediction computation is faster. - How would you evaluate the impact of reducing ETA computation time? - Distinguish between improvements in system latency and changes in ETA accuracy, and explain what the correct randomization unit should be in a marketplace setting. Your answer should explicitly discuss experimentation versus observational methods, spillover effects in two-sided marketplaces, and when cluster randomization is preferable to user-level randomization.