Brand revenue is not automatically Google Ads performance

Overall ROAS is a blended figure and can rise while you lose

A high overall ROAS says little about how Google Ads is performing, because above all it shows how much brand there is in the mix. My view: someone searching specifically for your shop is close to buying, and the click is usually cheap. That lifts the average, whatever the other campaigns are doing.

A worked example shows the consequence. Suppose a shop earns 40% on revenue before ad spend. Brand costs €5,000 a month and brings €100,000 of revenue, and this row stays the same in all three cases. Only the non-brand budget changes. The numbers are invented and rounded.

In the report, the cut looks like a success: ROAS rises from 7.2 to 9.7, yet €35,000 of revenue and €4,000 of contribution are missing. The expansion looks like a step backwards, although it leaves €4,000 more. Steering by overall ROAS therefore creates an incentive to shrink the business. Google itself points out that comparisons between campaign types can look different once you filter for brand Search campaigns. Part of the gap between Search and Shopping is then a brand effect, not a difference between the channels.

Attribution and effect are two different questions

The split tells you how much revenue runs through brand searches. It does not tell you whether that revenue would have been lost without the ad. Deriving the effect of Google Ads from subtracting brand replaces one fallacy with another.

There are two useful experiments on the effect, both involving large brands and neither with a link to the DACH region (Germany, Austria and Switzerland):

• eBay: in a field experiment, brand ads had no measurable short-term benefit. 99.5% of the click traffic that was lost came straight back through organic search. The authors extend this, with reservations, to well-known brands that buyers consider anyway.
• Bing, thousands of brands: without competitors on the brand term, the effect of brand ads on clicks was 1 to 4%, and lower for larger brands. Brands that did not bid while competitors bid on their name lost 18 to 42% of clicks. Defending the brand then had a clearly positive ROI.

My reading: brand revenue is neither worthless nor performance. Most of it is existing demand, and how much the ad adds depends mainly on who else is bidding on your brand. For a small shop without strong awareness I would not rely on the eBay figure; it is too far removed. There is therefore no flat discount factor, and a report should not contain one either.

Non-brand, incidentally, is not pure performance either. In a further test in the same study, new and infrequent users responded positively to ads, frequent buyers did not. The split makes the calculation more honest, but it proves no additional revenue.

Why subtracting the brand campaign is not enough

The usual method is: total revenue minus the brand campaign, done. That leaves every brand search that runs through Shopping, Performance Max or a generic campaign in non-brand, where it flatters the figure. A brand exclusion in PMax does not necessarily clear this up either: it applies to Search, Shopping and YouTube search inventory, and Shopping ads can optionally continue to serve on excluded brands.

On top of that, part of the picture always stays invisible. Search terms with too little query activity are left out of the report for privacy reasons. A list that adds up perfectly would therefore be suspicious.

How to split the report into three rows

1. Define the brand list. Shop name, own label, spelling variants, typos, brand plus product. The list stays the same from month to month, otherwise you are comparing definitions instead of results.
2. Collect brand. Pure brand campaigns plus brand terms from the search terms reports of all other campaigns, as far as your account shows spend and value per term there. In the PMax search terms report, segmenting by ad format helps, because it separates Shopping ads from text ads.
3. Collect non-brand. Only what you can prove: generic campaigns without brand terms, and the terms that clearly contain no brand.
4. Leave the remainder open. Everything that cannot be shown to belong to either row, above all revenue from PMax and Shopping without a search term assignment.

The example above then becomes this report:

The brand share is now a range: at least 56% (€100,000 of €180,000), at most 69% if the entire remainder were brand. The fact that the remainder, at 6.0, sits far above proven non-brand points to hidden brand. There is no benchmark for how high it should be. A young shop has little brand search, a well-known one a great deal, and both can be healthy. The figure only becomes meaningful over time.

The rule: is the range good enough for the decision?

The remainder does not have to go; it just must not tip the decision. For the lower bound of non-brand, add all of the remainder's spend and none of its revenue: €56,000 on €20,000, a ROAS of 2.8. For the upper bound, do the reverse: €80,000 on €16,000, a ROAS of 5.0. At a 40% margin the break-even ROAS is 2.5, below both ends: non-brand pays for itself however the remainder is distributed. That holds for the average. In the expansion case above, the ROAS of the additional budget is only 3.0.

At a 30% margin, break-even would be 3.3 and the result would be open. Then work on attribution pays off, and a properly configured brand exclusion in PMax is the first place to start. Before that, it would be busywork on a number that changes nothing.

For the budget question, non-brand is what counts from then on. I would treat brand as protection costs: an item with its own justification, whose effect only a test can show. The bidding decision itself is covered in bidding on your own brand. If you do not want to switch off completely, you can pause brand search in individual regions. How to evaluate something like that is covered in the article on incrementality.

Fair reporting means netting nothing off

The split is only fair if it also shows the uncomfortable parts. Four patterns make a monthly report useless:

• Overall ROAS stands alone as the success figure, without a split.
• Brand is removed, and its invisible share in PMax stays in non-brand. The remainder is missing.
• A corrected ROAS with a discount whose origin nobody can substantiate.
• The brand list changes quietly between two months.

Conversely, a large brand row is no criticism of whoever manages the account. Some brand searches may have been created by earlier ads; that is my assessment and cannot be measured in this calculation.

The split loses value when there is hardly any brand search: then overall ROAS is almost the same as non-brand, and the range changes no decision. Google features as of November 2026. For senior management, the three rows belong in the monthly report, alongside contribution after ad spend.