You're measuring a retail media campaign where some exposure is driven by platform delivery rules, so you cannot cleanly randomize every user. You still need a credible estimate of incremental sales from the campaign.
How would you design an incrementality test for a retail media campaign when you can't fully randomize users?