An incrementality testing technique where an algorithm bids on ad placements for a control group but deliberately doesn't win those placements, creating a matched control audience without actually exposing them to ads. It enables valid counterfactuals at scale without requiring a full holdout region.
Last updated:
July 28, 2026