A €5M DTC Brand Runs a 21-Day Geo Holdout on Google Brand: Worked Example
A worked example of a geo holdout on Google brand search: 10 paused DMAs, 10 matched controls, 21 days. Platform ROAS was 11.2x, the spillover-adjusted incremental ROAS was 1.45x, and the team cut brand spend by 40% instead of pausing it.
Quick answer
Pausing Google brand search in 10 matched US DMAs for 21 days cost the skincare brand an estimated $9,455 in revenue. Over the same period it saved $6,510 in ad spend. That works out to an incremental ROAS of 1.45x, compared with the 11.2x Google Ads reported. At a 55% contribution margin, brand search was losing money at full spend. The team cut it by 40% and kept bidding only in DMAs where competitors were bidding on the brand name.
Geo holdout on Google brand search
A geo holdout on brand search pauses branded ads in some regions and compares revenue against matched regions that keep running.
A geo holdout splits a market into comparable regions, such as US DMAs. Brand-keyword ads are switched off in the test regions and left running in the control regions. The control regions forecast what the test regions would have earned with ads on. The gap between that forecast and actual test-region revenue is your incremental revenue from brand search. It includes organic clicks that would have replaced paid clicks. Dividing it by the spend you saved gives incremental ROAS (iROAS), which is the number finance cares about.
This page walks through one real-shaped case: a €5M Shopify skincare brand that sells serums and SPF moisturisers, with about 45% of revenue coming from the US. The brand was spending $1,400 a day on Google brand search, roughly $42,000 a month.
Google Ads reported an 11.2x ROAS on the brand campaign. The new CFO asked a simple question: how much of that revenue would have come through the organic listing anyway? The Performance Manager ran the holdout to answer it.
Context: why brand search is the easiest place to overpay
Brand search catches demand that is already there. Someone who searched your brand name after a TikTok video or a Klaviyo email was already looking for you. Last-click attribution gives the paid ad full credit for that sale even when your organic result sits directly below it.
How incremental brand search is depends mostly on competitive pressure. If competitors bid on your brand name, the paid slot protects clicks you would otherwise lose. If no one bids on your terms, the organic result picks up most of the paused traffic.
Why the brand chose a geo split over a full pause
A national pause has no clean counterfactual, because seasonality, promotions and PR all move revenue at the same time. A geo split keeps a live control group exposed to the same calendar. That is why it survives finance review. For the full design trade-off, see Geo Split vs Full Pause: Designing a Brand-Paid Holdout That Survives Finance Review.
Test design: 10 paused DMAs vs 10 matched controls
The team ranked US DMAs by eight weeks of pre-period branded revenue, meaning Shopify orders from customers whose first session was a brand search or a direct visit. They grouped similar markets into pairs and randomly assigned one DMA from each pair to the test group. This produced 10 test DMAs and 10 control DMAs, with New York and Los Angeles left out because they were too large to pair.
In the pre-period, test-DMA revenue tracked control-DMA revenue at a stable ratio of 0.96, with a weekly correlation of 0.93. The test DMAs carried $310 a day of brand spend. Over 21 days, that meant $6,510 of spend would be held back.
The primary metric was total Shopify revenue in each DMA, not branded revenue only. This captured buyers who switched to the organic listing, direct visits or email. The team also froze promotions and Meta budgets by region during the test so nothing else could explain the difference.
Benchmark data: what the 21 days actually showed
21-day brand-search geo holdout results, €5M Shopify skincare brand (US, 20 DMAs)
| Metric | Value | How it was derived |
|---|---|---|
| Control DMA revenue (21 days) | $182,000 | Observed in Shopify |
| Expected test revenue | $174,720 | $182,000 × 0.96 pre-period ratio |
| Observed test revenue | $166,400 | Observed in Shopify |
| Raw revenue lost | $8,320 (−4.8%) | Expected minus observed |
| Geo leakage (ads still served in test DMAs) | 12% of normal impressions | From Google Ads geographic report |
| Spillover-adjusted lost revenue | $9,455 | $8,320 ÷ (1 − 0.12) |
| Spend saved | $6,510 | $310/day × 21 days |
| Incremental ROAS | 1.45x | $9,455 ÷ $6,510 |
| Platform-reported brand ROAS | 11.2x | Google Ads, last click |
| 90% interval on lost revenue | $2,100 – $16,800 | Placebo tests on pre-period weeks |
Revenue lost was only 4.8% of expected, but the spend was large compared with that revenue. The 11.2x platform ROAS overstated the true return by about 7.7 times, which is consistent with brands that have strong organic rankings and little competition on their own name.
Spillover adjustment and the decision the team made
Google's default location setting, which targets people in or interested in a location, kept serving about 12% of normal brand impressions inside the paused DMAs. Because the test was only partly switched off, the raw result understated the true effect. Dividing by 0.88 scaled it back up to what a full pause would show.
At a 55% contribution margin, $9,455 of incremental revenue produces about $5,200 of contribution. Spending $6,510 to earn that loses about $1,310 every three weeks. The interval was wide, but even its upper end gives only 1.4x in contribution terms, which barely covers the spend.
The team did not pause brand search everywhere. Auction Insights showed two competing serum brands bidding on the brand name in 6 of the 20 DMAs, and in those markets the revenue lost was about twice the average. The team kept full bids in those DMAs and switched the others to a tight target-impression-share cap, which cut brand spend by 40%. That freed roughly $17,000 a month for prospecting.
Testing ideas for your next brand holdout
Re-test the reduced setup after 60 days to confirm that total revenue held steady. Then run a second holdout that splits markets by competitor presence rather than at random, so you can price the cost of defending against conquesting directly.
Set the Google location option to 'presence' before launch to reduce leakage. Size the test with proper statistical analysis: with a 4–5% expected effect, 21 days across 20 DMAs was the minimum needed to get an interval that excludes zero.
Brand search geo holdout FAQ
Run it for at least 14–21 days so you cover full weekly cycles and your typical consideration window. For a skincare brand with an average order value around $60, 21 days was enough. Higher-priced products need longer.
Use at least 8–10 test and 8–10 control DMAs with stable pre-period revenue. Fewer markets widen the confidence interval until the result becomes useless. Leave out outlier metros that have no comparable partner.
Measure total revenue by region. When a paid click disappears, the customer often switches to organic search, direct or email, and only total revenue captures that switch.
Geo spillover is ad delivery or customer behaviour that crosses test boundaries, such as ads served to commuters or people Google classes as 'interested in' a location. It weakens the measured effect, so divide the raw result by the share of delivery you actually turned off.
Brand search is profitable when iROAS multiplied by contribution margin is above 1. At a 55% margin, you need an iROAS of roughly 1.8x or more. Results of 1–3x are common for brands without competitors bidding on their name.
Last-click attribution gives the brand ad full credit for buyers who would have clicked your organic result anyway. That inflation is the main reason to run a holdout.
Only where competitors bid on your terms. Check Auction Insights by region, and keep bidding where conquesting is active, as this brand did in 6 DMAs.
Yes, as long as orders can be tagged by shipping region or session geography and compared with a matched control group. The hard part is building the pre-period baseline, and imported historical GA4 data makes that much easier.
A geo split gives you a live counterfactual and a smaller revenue risk, so finance teams trust it more. A full pause is simpler but gets mixed up with seasonality and promotions.
Run placebo tests: apply the same analysis to pre-period weeks when nothing changed and look at how much the gap varies. Synthetic control or difference-in-differences methods formalise this approach.
Track CAC, channels, and funnel conversion in one place
Metricuno connects ad spend, funnel events, and revenue so you can see CAC by channel, cohort, and campaign — without stitching together five tools.