Incrementality
Measure outcomes that would not have occurred without the campaign.
Quick learn
A fictional beauty retailer randomly assigns 20,000 eligible users to ads and 10,000 to holdout. The treatment group produces 1,200 purchases, and holdout produces 500. Campaign spend is ₹60,000.
Deep dive
Normalize purchase counts by the assigned group size. The holdout estimates baseline demand. Use assignment as the analysis basis even if some treatment users do not receive an impression. Check for spillovers and uncertainty.
Worked decision
Treatment buys at 6%, holdout at 5%. At treatment scale, the 1 percentage point difference implies 200 incremental purchases. Incremental cost per purchase is ₹60,000 ÷ 200 = ₹300.
When the decision changes
If each incremental order contributes ₹250, spend exceeds incremental contribution by ₹10,000. A platform can claim attributed orders while the holdout estimates fewer causal orders.