Separate brand from non-brand before you trust any other number in the account. Not because blended reporting inflates ROAS (everyone knows that) but because blending two streams with opposite economics into one average makes every figure downstream of it uninterpretable, including the ones you would use to diagnose the problem. It is not a distortion you can mentally correct for. It is a loss of resolution.
And it has a property that makes it genuinely treacherous rather than merely wrong: the error grows as your business succeeds. Brand search volume rises with awareness, and awareness rises from press, email, organic, word of mouth, a product that works. All of that lands in your paid dashboard as improving efficiency. The better your non-paid marketing performs, the more your paid reporting flatters itself, and the more confidently you will scale the part of the account that is merely collecting the credit.
The decomposition, worked
Here is what a blended figure conceals. Round numbers, ordinary proportions: a mid-market account where brand is a small share of spend and a large share of reported revenue:
AS REPORTED
spend 50,000
revenue 200,000
ROAS 4.00x
SPLIT BY INTENT
brand spend 8,000 rev 88,000 -> 11.00x
non-brand spend 42,000 rev 112,000 -> 2.67x
brand = 16% of spend, 44% of revenue
NOW STRIP THE NON-INCREMENTAL BRAND REVENUE
assume 85% of brand revenue would have
arrived via the organic listing anyway
incremental brand revenue 13,200
brand incremental ROAS 1.65x
total incremental revenue 125,200
ACCOUNT INCREMENTAL ROAS 2.50x
AGAINST A 3.0x TARGET
reported 4.00x -> scale
non-brand 2.67x -> tighten
incremental 2.50x -> restructureThe reported figure is sixty percent higher than the incremental one, and the three views prescribe three different actions against the same target. Notice which one you would act on if nobody had done this arithmetic: the account looks like it is beating target by a third and asking for more budget.
If you cannot state your non-brand ROAS in five seconds, cleanly separated, the account is not telling you where growth comes from. And the eighty-five percent assumption above is exactly that, an assumption. The only way to replace it with a measurement is a brand-search shutoff test, which is the cheapest experiment in paid media.
The real damage is done by the bidding algorithm
The reporting problem is the visible half. The mechanical problem is worse, and it is what turns a bad number into a deteriorating account.
Smart Bidding optimizes toward value at a target, and it does not care which intent produced the value. Put an eleven-times stream and a two-and-a-half-times stream under a single four-times target and the algorithm can hit that target comfortably by leaning on brand, which means it will bid non-brand down below where it would have bid if non-brand were being measured on its own. The blended target is achievable for the portfolio and unachievable for the growth engine inside it, so the account quietly converges toward harvesting demand it already had.
Reported efficiency holds or improves throughout. Incremental revenue flattens. That is the specific mechanism behind the pattern where ad performance keeps looking better while margin keeps shrinking, and it is invisible unless the streams are split.
The corollary is a trap on the way out. Cut brand spend and blended ROAS falls, because you removed the cheapest revenue in the account. It looks like the cut caused damage. It did not. It removed a subsidy from the average. Anyone reading a blended dashboard will conclude the opposite and put the spend back.
Separation leaks, and you have to measure the leak
The standard advice is to split brand into its own campaign and add brand negatives to non-brand. Necessary, and nowhere near sufficient: brand traffic finds its way back in through at least six routes, several of which no negative keyword list can close:
- Close variants: Google matches terms with the same meaning, so a brand negative will not reliably stop every brand-adjacent query, particularly misspellings and spacing variants of the brand itself.
- Performance Max reabsorption: Left unexcluded, PMax will happily serve on your brand queries and report them inside a blended average with no keyword to point at. Brand exclusions and account-level negatives are both required.
- Shopping campaigns: They match on the feed rather than on keywords, so brand queries land in them with no keyword-level control at all. Campaign-level negatives help; query-level reporting is the only real defense.
- Brand-plus-modifier queries in broad match: “Brandname reviews”, “brandname discount code”, “is brandname any good” are brand intent wearing research clothing, and broad match in a non-brand campaign will collect them.
- Comparison queries: “Brandname vs competitor” is genuinely mixed (the searcher knows you and is deciding) and it will sit in whichever campaign happens to match it first.
- Partner and expansion inventory: Search partners and display expansion introduce traffic your keyword-level segmentation never fully describes.
Because it leaks, treat separation as a measured quantity rather than a configuration you completed. Segment your non-brand campaigns by search term and compute what share of their spend and revenue came from queries containing brand tokens. That leak rate belongs in the monthly report next to non-brand ROAS, and it should be trending toward zero rather than assumed to be there. An account with a fifteen percent brand leak in its non-brand campaigns is still reporting a materially inflated non-brand figure, however clean the campaign names look.
What to do with brand once it is separated
Separating brand is not an argument for switching it off, and the decomposition above is not evidence that brand spend is waste. Brand campaigns defend against competitors bidding on your name, control the message above the organic listing, and capture demand cheaply, and a 1.65-times incremental return can be perfectly good value on a small budget. The point is to judge it as what it is: a defensive demand-capture line with a modest incremental return and a low ceiling, funded deliberately rather than by an algorithm that mistook it for your growth engine.
Practically that means separate budgets, separate targets set from each stream’s own economics, and two lines in every report that are never blended into one headline. Non-brand carries the growth question and should be scaled against the margin-derived target it can actually sustain. Brand gets a cap, not a target.
Where the split is genuinely ambiguous
- Brands whose name is also a category term: A retailer literally called something like Rugs Outlet has no clean boundary: the same query is brand and category simultaneously, and any split you choose is a convention rather than a fact. Document the convention and keep it stable, because switching it mid-year makes your own trend data useless.
- New brands with no awareness base: If nobody knows you, a brand search often happened because an ad created it: someone saw a Meta ad, remembered the name, and searched. There the brand click is closer to genuinely incremental, and an eighty-five percent discount would understate its value badly. The assumption scales with awareness.
- Comparison and review intent: “Brand vs competitor” and “brand reviews” are decision-stage queries where the ad may well change the outcome. Filing them as pure brand capture is convenient and probably wrong.
- Separation costs some algorithmic efficiency: Splitting campaigns splits the conversion data Smart Bidding learns from, and brand data is high-volume and clean. On small accounts the loss is real, which is the honest argument for reporting-level separation before structural separation when volume is thin.
- The 85% figure is our working assumption, not a constant: It varies by category, by organic strength, by how prominent your organic listing is, and by competitor behavior on your terms. Anyone quoting you a fixed number for this has not tested it. Measure yours with a brand-search shutoff and then use your number.
- None of this survives a bad revenue source: Splitting brand from non-brand on platform-reported conversion value corrects one bias while leaving a larger one in place. The source question comes first. See MER vs ROAS.
The first time a brand sees its true non-brand return, the number is usually sobering. It is also the most useful figure they have looked at in months, because for the first time it describes the part of the account that creates customers rather than the part that collects the ones they already had. Every scaling decision that matters is made on that number, and until the split exists, nobody in the room has ever seen it.
Common questions