The reliable way to measure it is a controlled test: stop or change the advertising for one group, keep it running for a comparable group, and compare what each group buys. Every attribution report, however detailed, is an inference.
Why attribution is not enough
Attribution shares out the credit for sales that happened among the ads a customer saw. It cannot tell you which of those sales would have happened with no ad at all. That matters most where advertising reaches people who were already on their way to buy: searches for your own brand, and ads shown to recent visitors.
Three ways to test
- A regional holdout. Switch the advertising off, or cut it, in a set of regions that normally behave like the rest, and compare their sales with the regions where it kept running, against the same weeks before the test.
- A time-based test. Pause a channel for a set period and compare sales with the matched period a year earlier. It needs no tools, and it is the weakest of the three, because anything else that changed in that period lands in the result.
- A platform lift study. Google Ads and Meta both offer conversion lift studies, which hold back a random group of people from seeing the ads and compare the two groups. Each platform sets its own eligibility, usually tied to spend and conversion volume.
Designing a fair test
- Measure the result in your own sales or CRM data, not in the platform's report.
- Choose groups that behaved alike before the test, and check that they did.
- Run it for at least one full buying cycle, so slow decisions are counted.
- Decide before you start what result would change the budget, so the answer is not argued after the fact.
- Change nothing else in the test groups while it runs.
Reading the result
The extra sales in the group that saw the advertising, compared with the group that did not, are the incremental sales. Multiply them by your gross margin and set the result against what the advertising cost in those regions or that period. That is the return the advertising actually produced, and it can be a very different number from the one in the platform's report.
Where to start
- Paid search on your own brand name.
- Remarketing to people who have already visited.
- Any channel whose reported sales finance cannot find in the books.