SlackThe idea
Withhold ads from a comparable group and compare orders with the group that saw them. The difference is the lift the ads caused.
Design before you start
- Split: by region or by audience, in a way that keeps the groups similar in size and in past sales.
- Period: long enough to collect enough orders, and the same for both groups.
- Metric: store orders or revenue, not platform reported conversions.
- Decision rule: what lift means keep, cut or retest, written before the test.
While it runs
- Do not change budgets, offers or creative in one group only.
- Note anything else that could differ: a sale, a stock out, a press mention.
Reading the result
- Compare the groups on the store's own numbers.
- A small difference on a small sample proves little. Say so.
- Repeat if the decision is expensive.
The attribution windows guide explains why a window alone cannot prove cause.
