To know whether ads generate sales that would not exist without them, you do not need a laboratory. Simple, well-planned geo and holdout tests can give useful answers even on small budgets, provided some uncertainty is accepted.
The essentials in a definition
Incrementality without a lab is measuring the real effect of ads with simple designs: geo tests (regions with and without ads) and holdout tests (groups that do not see the ads). They work with small budgets if planned, with comparable regions, a defined duration and one change at a time. They complement conversion measurement and models such as Meridian.
A geo test in simple terms
You choose two similar areas (by population, sales, seasonality). In one you keep spending. In the other you reduce or pause it for a few weeks. You compare how sales evolve in both.
If sales in the paused area fall much more, the ads were incremental. If little changes, the effect is smaller than reports suggest.
A holdout test
Instead of areas, you exclude part of the audience (for example 10 to 20%) from the ads and compare behaviour with the exposed group. Some platforms offer this function. Manual versions can also be designed with lists.
It is especially useful for remarketing and brand campaigns, where the risk of paying for people who would buy anyway is higher.
A simple incrementality test
- Choose two comparable areas or groups
- Reduce or pause ads in one of them
- Compare sales over the defined period
- Read the result as a range and decide
Design care
Choose comparable areas or groups. Define the duration before starting (generally a few weeks, covering the buying cycle). Avoid changing other things in the same period. Record promotions and external events.
Measure outcome in sales or qualified leads, not just attributed conversions.
Pausing in one area for a month can teach more than a year of reports.
Reading and acting
Think in ranges rather than exact numbers. A test with little data says “probably between X and Y”. That already helps: if the lower bound still justifies the investment, go ahead. If the upper bound falls below cost, cut.
Repeat when conditions change. A result from a year ago may no longer apply.
Decision table
| Design | How to do it | Limit |
|---|---|---|
| Geo test | Pause or change ads in a few similar regions | Needs comparable regions and time |
| Audience holdout test | Exclude a random part of remarketing | Depends on platform tools |
| Before and after | Compare comparable periods | Weak: external factors |
| Temporary pause | Pause a channel for 2 to 4 weeks | Risks sales; plan carefully |
Common mistakes to avoid
- Choosing very different regions. Without comparability, the difference may be due to the region.
- Testing during promotions. External factors distort the effect; choose a stable period.
- Stopping early. Insufficient duration produces false positives.
How to measure the result
Define the main metric (sales, leads or margin), calculate the difference between regions before and after the test, and record the uncertainty. Repeat if the conclusion is critical for budget decisions.
Limits and exceptions
Geo and holdout tests are imperfect: regions are never identical, there is leakage between zones (people travelling, ads crossing borders) and local factors can weigh. In very small businesses or with few regions, statistical power is limited. Interpret results with caution, repeat when possible and combine with other evidence before changing large budgets.
Illustrative example
Illustrative example: a brand with similar sales in four districts pauses remarketing in two for four weeks and keeps it in the other two. The sales difference is small and within usual variation, and the brand reduces its remarketing budget, reinvesting in acquisition channels. The conclusion was documented so the test can be repeated in six months. Studies such as those from Think with Google help formulate hypotheses.
Frequently asked questions
Do I need a big budget?
No. Simple regional or holdout tests work with modest budgets.
How long should it run?
Long enough to cover the buying cycle, generally a few weeks.
What if the business is small?
Use larger groups and longer periods, and accept greater uncertainty.
How many regions are needed?
Ideally several comparable pairs; with few points, interpret the result cautiously and as an indication.
Is pausing ads risky?
It may cost sales in the short term; limit the test to small regions and a short period, and measure the effect.
Geo and holdout tests let you measure the incremental effect of ads on small budgets, if there are comparable regions, a defined duration and a single change.
Sources and recommended reading
Last editorial review: October 2026. Informational content; always confirm current conditions in the official sources before deciding.