You work on a consumer goods campaign experiment and need to decide how to randomize the test. The campaign is delivered through retail or customer touchpoints where the same shopper may be exposed multiple times, and there is concern that the wrong unit could bias the results.
How would you choose the right unit of randomization for this campaign experiment, and what trade-offs would you consider if user-level, store-level, or geo-level randomization are all plausible?
