How to run a ghost-ads incrementality test on paid social

How to run a ghost-ads incrementality test on paid social

The question: platform-reported ROAS says your prospecting campaigns print money. Is that causal, or are the ads being shown to people who'd convert anyway? Ghost ads are the cleanest way to find out.

The playbook:

— Understand the mechanism. The ad system runs its auction as normal, then for a randomized control group it records who would have seen your ad and shows them nothing (or a placeholder). Test and control are matched on the platform's own targeting — eliminating the activity bias that plagued older PSA-based studies (Johnson, Lewis, Nubbemeyer documented this).
— Pre-commit to your primary metric and minimum detectable effect.
— Check you have the scale for power. Conversion-lift tests need large audiences; rare events demand huge ones.
— Run for full purchase cycles, not days. Short windows catch only impulse buyers and miss considered purchases.
— Read incremental conversions, not platform-attributed ones. The whole point is to ignore the platform's self-credit.

The nuance: lift studies estimate the average incremental effect for that audience and creative. It won't always generalize when you scale spend — saturation bends the curve. Re-test at the new budget.

Bottom line for practitioners: ghost/conversion-lift tests are the gold standard for paid social because randomization severs correlation from causation by design. Treat reported ROAS as a hypothesis and the lift test as the verdict.