You work on a retail media team measuring whether a campaign drove incremental sales, but you cannot cleanly randomize exposure at the user level because delivery and matching are imperfect. You still need a credible experiment design that can estimate lift while accounting for spillovers and operational bias.
How would you think about incrementality testing for this campaign when user-level randomization is not feasible? Walk through the experiment design you would choose, how you would measure impact, and what risks could bias the estimate.