04 / Testing
Incrementality testing: geo-lift and holdout
Every advertising platform reports the conversions it believes it caused. An incrementality test measures what would have happened if you had not run the advertising at all. The gap between those two numbers is the only one worth budgeting against.
Test parameters
- Design
- geo-lift or holdout
- Cells
- matched pairs
- Power
- set before launch
- Read-out
- lift and interval
Test regions against matched control regions. The shaded area is the incremental effect: what the advertising caused, rather than what it was present for.
Why this is worth doing now
On 20 May 2026 Google reduced the minimum spend for incrementality experiments from roughly one hundred thousand dollars to roughly five thousand. Testing that was only available to enterprise advertisers is now within reach of businesses spending a fraction of that, and most of them have not noticed yet.
The reason it matters is that platform-reported performance and measured incremental performance routinely differ by a large margin, in the direction that flatters the platform. Brand search is the clearest example: much of what it is credited with would have happened anyway. Until a business has measured that for itself, every budget decision rests on a number supplied by the party being paid.
How a test runs
-
Design
Which channel, which regions, and how long. Test and control regions matched on the things that actually predict sales rather than on population alone.
-
Power
The minimum effect the test can detect, calculated before anything is switched off. A test that cannot detect the effect you care about is worse than no test.
-
Run
Spend held or withdrawn in the control cells, everything else held constant, and the test monitored for the things that invalidate it.
-
Read-out
Measured lift with a confidence interval, what it implies for the channel, and a plain statement of what the test cannot tell you.
A geo test reads out against your own conversion data, so server-side tagging and consent has to be sound before a test is worth running. For effects small enough to detect on a page, use conversion rate optimisation.
Find out whether a test would read at your spend
One email starts it. No form, no queue, no account to create, and no discovery call before anyone has told you anything useful.
That opens a message to info@mktg-consulting.com with the four things we need already listed. Answer what you can and we will ask for the rest.
What to include
- Which of the four areas you are asking about, or that you are not sure
- What is already in place, and roughly how long it has been there
- What is driving the timing, because that usually decides the shape of the work
We reply to every enquiry within one working day, including the ones we are not right for. If it is not work we should be doing we will say so and, where we can, say who should. A site promising a reply and never saying when is a promise nobody can hold us to.