Incrementality Testing for Mid-Size Brands: <blue>Geo Holdouts and Conversion Lift</blue>

Incrementality testing shows how many sales your ads actually caused, by comparing a group that saw the ads with a similar group that did not. For mid-size brands, the two most practical methods are geo holdouts, where you turn a channel off in selected regions and compare results with matched regions, and conversion lift studies run by the ad platforms. Neither needs an enterprise budget. They need a clear question, enough conversion volume, and backend sales data you trust.

Key takeaways

• Platform-reported ROAS measures credit, not cause. Incrementality tests measure cause.
• Geo holdouts work across any channel and use your own sales data, which makes them a strong fit for mid-size brands.
• Conversion lift studies from Google and Meta are useful but usually need an account representative and enough volume to qualify.
• Use test results to calibrate how much you trust platform numbers, not as a one-time verdict.

Why mid-size brands need incrementality testing

Every ad platform reports the conversions it believes it influenced, and each one credits itself. Add up the revenue claimed by Meta, Google, and TikTok and it often exceeds what your store recorded. Retargeting and branded search can look especially strong because they reach people who were already likely to buy. Teams running margin-focused paid media use incrementality tests to find out which spend is creating sales and which is collecting credit.

Mid-size brands feel this most. Budgets are big enough that misallocation is expensive, but rarely big enough for a full marketing mix model with years of clean data. A few well-designed tests a year close that gap.

Where incrementality fits

Think of measurement as a ladder. Engagement metrics show ads are being seen. Platform conversions show which campaigns get credit. Backend data shows whether the business actually grew. Incrementality sits at the top, because it is the only level that shows what would have happened without the ads.

How geo holdouts work

A geo holdout splits your market into regions, such as states, DMAs, or postal areas. In the test regions, a channel or campaign is paused or changed. In the control regions, it keeps running. Because the regions were chosen to behave alike before the test, the difference in sales during the test is a reasonable estimate of what the channel contributes.

1. Pick one question. For example, what does Meta prospecting add beyond everything else we run?
2. Choose matched regions. Use historical sales to find test and control regions that move together week to week. Meta's open-source GeoLift package includes tools for market selection and power analysis.
3. Size the holdout. Hold out enough sales volume to detect a realistic effect, but not so much that the test becomes expensive.
4. Run long enough. Cover at least one full purchase cycle and avoid major promotions or launches that hit regions unevenly.
5. Measure in your backend. Compare orders, new customers, and revenue from your own systems, not platform reports.
6. Calculate incremental return. Divide incremental revenue, or better, incremental contribution margin, by the spend in the test regions.

Hypothetical example: A brand pauses Meta prospecting in regions that normally produce about 15% of sales. Over four weeks, those regions come in $36,000 below what the matched control regions predicted. The paused spend would have been $24,000, so incremental ROAS is 1.5. Meta's in-platform ROAS for the same regions was 3.0. The channel is working, but it creates about half the revenue the platform credits it with, and budget decisions should reflect that.

How conversion lift studies work

Conversion lift studies are run inside the ad platforms. The platform randomly withholds ads from a control group and compares conversions with the group that saw them.

As of November 2026, Google offers two types. Users-based Conversion Lift splits groups by aggregated user attributes and reports results by conversion action, age, and gender. Geography-based Conversion Lift splits by region, supports offline data, and covers Search, Shopping, Performance Max, Demand Gen, Video, Display, and App campaigns. Google's lift types comparison page notes that geo studies tend to need higher budgets, and access generally goes through a Google account representative. Meta offers conversion lift studies as well, typically set up with a Meta representative.

Platform lift studies are convenient and statistically rigorous within one platform. Their limits are that they measure one channel at a time, they depend on the platform's own conversion tracking, and they may not be available below certain spend levels.

Geo holdouts vs. conversion lift

• Channels covered. Geo holdouts: any channel, including offline. Conversion lift: one platform per study.
• Data source. Geo holdouts: your backend sales. Conversion lift: platform-tracked conversions.
• Access. Geo holdouts: run them yourself. Conversion lift: usually through a platform representative.
• Precision. Geo holdouts: noisier, need careful region matching. Conversion lift: more precise within the platform.

Common mistakes

• Testing with too little volume. Small tests produce wide ranges that cannot support a decision.
• Ending too early. Short tests miss delayed purchases and overreact to noise.
• Measuring revenue instead of margin. A channel can lift revenue and still lose money. The comparison of contribution margin vs. ROAS explains why.
• Treating one test as permanent. Results drift as creative, audiences, and spend change. Retest your biggest channels periodically.

Turning results into decisions

The most useful output of a test is a calibration factor. If Meta's true incremental return is about half its reported ROAS, you can adjust targets and keep using platform data for daily optimization. That only works when ad spend, orders, and costs live in one place, which is the role of unified marketing data. It is also central to how Onward turns fragmented metrics into decisions.

Frequently asked questions

How much do I need to spend to run an incrementality test?

There is no single threshold. What matters is whether the holdout covers enough conversions to detect a realistic lift. A power analysis before the test tells you whether your volume is sufficient. Platform lift studies may also have their own minimums.

How long should a geo holdout run?

Long enough to cover at least one full purchase cycle and collect sufficient conversions, which is often several weeks. Brands with longer consideration periods need longer tests.

Will a holdout hurt sales?

Some sales may be lost in the holdout regions while the test runs. Keeping the holdout share modest limits the cost, and the insight usually pays for it many times over in better budget allocation.

The bottom line

Incrementality testing is how a mid-size brand learns what its ads really cause. Geo holdouts give a cross-channel view grounded in your own sales data, and conversion lift studies add precision inside each platform. Run a few a year, measure in margin, and use the results to calibrate everything else.

If you want help designing your first incrementality test, talk to Onward's team.