Design an experiment for thermal bags

We intend to equip couriers with insulated bags to cut the cost of refunds caused by cold food. Design a rigorous experiment: (a) Select the randomization unit—courier, store, or zone—and justify your choice while accounting for network effects and spillover/contamination, such as couriers operating across multiple zones or stores assigned to both treatment and control. (b) Define the primary metric (refund cost per order due to cold food) and guardrail metrics (delivery ETA accuracy, contact rate, reorder rate, courier supply hours). (c) Specify stratified or clustered randomization by city and peak versus off-peak periods, and explain how to handle seasonality and holidays, such as staggered rollouts or difference-in-differences on pre-period trends. (d) Address noncompliance (e.g., couriers not using the bags) and treatment measurement (telemetry or audit photos), and propose an encouragement design or instrumental variables approach to estimate the local average treatment effect. (e) Provide the analysis plan, including CUPED or covariate adjustment (distance, cuisine, temperature, store), heterogeneity analyses by cuisine and distance deciles, cluster-robust standard errors for variance estimation, sequential monitoring rules, and pre-registered success thresholds.

Overview: This question assesses a data scientist's experimental design and causal inference abilities: choosing the randomization unit, addressing contamination and clustering, specifying outcomes and guardrails, measuring treatment compliance, and performing statistical analysis of heterogeneity and variance.