A design that cannot resolve the question is declined rather than run.
A holdout suppresses delivery across a matched share of geography or audience and compares outcomes with the exposed share over a fixed window. Before a design is offered, the experiment service checks whether the account has the conversion volume, the run time and the operational room for it to say anything; when it does not, it declines instead of returning a number with no power behind it. Results are stated as estimates, with design, window, matched criteria and confidence attached. Incrementality never overwrites attribution credit — both stay on the record.
Designed only where power exists
Geo splits need enough regions and enough conversions per region to detect a difference; audience holdouts need a suppressible share that does not distort delivery elsewhere. The feasibility check runs on the account's own history, and a design that cannot resolve the question is declined with the reason stated.
Matched controls locked for the window
Control regions are matched on prior conversion rate and population, then frozen, and the campaigns involved are protected from routine budget movement for the run. Ending a holdout early restores the prior delivery state, because the suppression was written as a reversible action in the first place.
Credit and lift side by side
The read shows what attribution credits the channel with and, separately, the estimated difference between exposed and held-out populations. A wide gap between the two is not an error in either figure — it is the measure of how much credited volume was going to arrive regardless.
In this capability
- Geo holdouts
- Audience holdouts
- Power and feasibility checks
- Separate incrementality record
Experiments
The rest of Experiments
Creative Tests
Concept, copy, visual and format tests that report at concept level, not at the ad ID.
Audience Tests
Expansion, suppression and seed tests, graded on which segments ended up as paying customers.
Channel Tests
Budget allocation and market-entry tests: what a channel is worth at a different spend level, or at all.
Landing-Page Tests
Message and page variants joined to outcomes through content IDs and your web analytics.
Learning Agenda
A ranked queue of open business questions, so testing is a programme rather than an occasional habit.
One goal. Every channel. Real business truth.
Connect your ad accounts, say what a good customer is worth, and let the Growth Brain work the objective inside limits you set.
No ad spend commitment · Connect one channel to start · [email protected]