Audience deliberately withheld from a campaign to measure uplift.
Standard in Meta Conversion Lift and Google Geo Experiments. Without a holdout, you cannot distinguish correlation from causation.