The 30-Minute Branded Bleed Audit

• Pull PMax search category insights from the Insights tab (top search categories and themes) instead of hunting for a row-level search terms report. Standard Performance Max does not expose one the way a Search campaign does.
• Tag every category and theme as brand or non-brand using a list built the same way as a Search brand list: exact name, variants, common misspellings pulled from Search Console, and brand-plus-product combinations.
• Estimate branded share of conversions from the tagged categories, then sanity-check it against the branded share already visible in a parallel Search campaign, if one exists, since the two channels usually track close to each other.
• Compute a rough branded-adjusted ROAS: reported ROAS multiplied by the non-branded share of conversions. This is an estimate, not a precise figure, which is exactly why the holdout in section 2 exists.
• Repeat the same category tagging for AI Max reporting, since it inherits the same branded-demand exposure that PMax carries.
• Flag the result with a confidence level, high, medium, or low, based on how cleanly the categories separated. A bad answer is reporting the branded-adjusted number as fact when the category tagging was mostly guesswork.

2. Designing a geo holdout with no data team

1. 01

   Pick matched regions, not an arbitrary split

   Group the account's sales geography into two sets of regions matched on size and historical revenue share, close enough that swapping which set is the holdout wouldn't change the story. States, provinces, or DMAs work depending on the market. A single huge region against a handful of small ones does not.

2. 02

   Pause paid spend, including brand, in the holdout regions only

   Turn off Search, PMax, and AI Max serving into the holdout regions using location exclusions at the campaign level. Leave the control regions running exactly as they were. The holdout regions keep organic, email, and any other channel untouched. Only paid media stops.

3. 03

   Hold the split for a full purchase cycle plus a buffer

   Most ecom categories need two to four weeks to smooth out day-of-week and payday noise. Considered-purchase categories need longer. Do not end the test early because the first week looks dramatic in either direction; early weeks are the noisiest.

4. 04

   Compare total revenue per region, not ad-attributed revenue alone

   Pull actual order revenue by shipping region from the store platform for both sets of regions, for the holdout window and for the same window last cycle. The gap between the holdout regions' revenue drop and the control regions' revenue is the closest a manual test gets to a true incrementality read.

Reported ROAS versus the incremental number, two composite examples

Composite, illustrative figures from Ad-Lab audits, not one specific client's account. The math: incremental ROAS equals reported ROAS multiplied by the share that was not already-owned brand demand.

• The number that matters is total company revenue during the holdout window against forecast, not Performance Max's own reported revenue during the same window.
• A channel can report a revenue drop in the holdout regions while total regional revenue barely moves, because the sale still happened through another channel. That outcome confirms the test is working as intended.
• Seasonality and promotions are checked against the same window last cycle before attributing any gap to the holdout itself. A site-wide sale landing inside the test window will contaminate the read.
• The result is expressed as a range, not a single decimal, given how much a manual holdout depends on regions that are close but never perfectly matched.
• The result holds up under a second look: rerun the same window calculation a week later once any lagged conversions have settled, since a geo holdout reads cleaner after the reporting lag closes.
• The finding gets written down with the date range and the regions used, so the next person running this audit isn't starting from zero.

• Competitors or resellers are bidding on the brand's own name, and losing that top slot costs more in diverted sales than the branded campaign costs to run.
• The category has enough marketplace or comparison-shopping presence that a blank paid slot next to the brand name reads as absence, not confidence.
• The branded campaign budget is small enough relative to total spend that even a full bleed adjustment barely moves the account's blended efficiency.
• The business is entering a launch or PR window where branded search volume is about to spike from earned coverage, and paid presence is deliberate insurance during that spike.
• The account is too small for a clean holdout yet, too few regions or too little baseline revenue to separate a real gap from noise, and the honest answer is to wait for enough scale rather than force a read on too little data.

Automated incrementality tooling as the shortcut

External resources

Authoritative references we link to alongside the template. Read them before running the audit.

• Google Ads, conversion lift and incrementalityGoogle's own geo-experiment and lift-measurement tooling, the automated version of section 2.
• Google Ads, brand exclusions in Performance MaxHow to act on the result once the branded share is known.
• Google Ads, search-terms reportReference for the Search-side brand list this audit's category tagging borrows from.
• Think with Google, measuring incrementalityBackground reading on incrementality measurement approaches beyond this audit's manual method.