After managing hundreds of brands on Amazon, we can tell you which campaign gets shown first on every monthly call. It is the branded keyword campaign, and it is shown first because it reports the best number in the account. Eight, ten, twelve times return on ad spend. Everyone nods. Nobody asks the only question that matters about branded spend, which is how much of that revenue would have happened if the campaign did not exist.
In May we published a case for defending branded search through Sponsored Brands, and that advice stands. What that post did not do, and what almost nobody does, is tell you how to find out what the defense is actually buying. The ROAS on a branded campaign is the least informative number in your ad console, because it is mostly measuring demand you created somewhere else, and the report has no column for “would have bought anyway.”
This is how to build that column yourself, in September, before conquest bidding and gift traffic change the answer.
Branded spend buys two things and only one of them is incremental
When a shopper types your brand name and clicks your ad, one of two things happened.
Defense. A competitor’s card would have sat above your organic result, and some share of those shoppers would have taken it. Your ad occupied the slot instead. That sale is incremental to the ad, because without it a percentage of that traffic leaks.
Capture. Nobody was conquesting the query, your organic result sat right there, and the shopper would have clicked it. Your ad intercepted a sale that was already yours, and attributed it to advertising. You still made the sale. You paid a click for it.
Every branded campaign is a blend of the two, and the report shows them as one number. The blend depends on who is bidding on your name, how strong your organic position is on your own brand terms, and how loyal the shopper typing the query actually is. All three of those move in Q4, which is why the number you measure in September is the one you need in November.
The math that the ROAS hides
Take a $200K/month brand. Branded campaign spend is $2,400 a month, attributed sales are $26,000. That is 10.8x ROAS, a 9% ACOS, and the best-looking line in the deck.
Now suppose you turn it off and 85% of those purchases still happen, because your organic result is first on your own brand name and nobody serious is conquesting you. Incremental revenue is $3,900. Incremental ROAS is 1.6x. At a 55% contribution margin that is roughly $2,150 of contribution for $2,400 of spend. The best-looking campaign in the account is losing money.
Change one number. Suppose only 60% returns organically, because two competitors are running Sponsored Brands on your name and the headline slot is theirs the moment you leave. Incremental revenue is $10,400, incremental ROAS is 4.3x, and the campaign is clearly worth funding, probably at a higher bid than it currently gets.
Same campaign, same 10.8x on the report. The decision flips entirely on a number you have never measured: the share of branded purchases that return organically when the ad is gone. We are not going to tell you what that number is for your account, because nobody can. Third-party writeups circulate figures like 80% to 90% returning organically, and those are somebody else’s categories. The only version that applies to you is the one you measure.
Why the blended ACOS keeps this hidden
There is a second, quieter cost. Branded sales sit inside your total ad-attributed sales and drag the blended ACOS down.
An account with 40% of attributed sales on branded terms at 9% ACOS and 60% on non-branded at 38% reports a blended ACOS around 26%. That is the number in the monthly report, that is the number the agency is graded on, and it looks fine. Strip out the branded half and the account’s actual acquisition efficiency is 38%, which is a different conversation.
This is not an accusation. It is an incentive. An agency graded on blended ACOS has a structural reason to keep branded campaigns fully funded, because branded spend makes the graded number better regardless of whether it is incremental. If your agency has never volunteered a branded versus non-branded split, ask for one this week. A shop that runs accounts properly produces it in minutes.
How to run a holdout on Amazon, since you cannot geo-test
Most incrementality advice circulating for Amazon is imported from Google and Meta, where you can hold out a geography. Sponsored Products and Sponsored Brands in the US do not offer geographic targeting. A “geo-holdout” on Amazon sponsored ads is not a thing you can configure, so the test has to be built on time or on SKUs.
The time-based holdout. Pause branded campaigns for two weeks, run them for two weeks, repeat once. Four weeks minimum, eight is better. Read the result on total purchases for your branded queries, not on attributed sales.
The SKU-split holdout. If you have sibling ASINs with similar branded query volume, hold branded ads out on half of them and leave the other half running. Cleaner against seasonality, weaker if the SKUs are not comparable.
What you measure. Search Query Performance reports total purchases on each query, whether or not an ad was involved. Pull it weekly for your branded query set. That is the number that tells you whether sales held. Attributed sales will collapse when the campaign is paused, by construction, and they tell you nothing. Add Business Reports sessions and units for the ASINs as a cross-check.
The arithmetic. Average weekly branded purchases with ads on, minus average weekly branded purchases with ads off, divided by the ads-on number. That is the share of branded purchases that was incremental. Everything else was capture.
The confounds. Log every deal, coupon, stockout, price change, and competitor launch across the window with a date. Read the ads-on weeks only after attribution has settled, seven days for Sponsored Products and fourteen for Sponsored Brands, or you will compare a finished week against an unfinished one and conclude the wrong thing.
What to leave running. If you want to know what the Sponsored Brands headline is worth separately from Sponsored Products, hold them out one at a time. Most brands should test the Sponsored Products branded campaign first, because it is usually the larger spend and the more likely capture.
Reading the three results
Branded purchases drop less than 10% with ads off. You have a capture campaign. Your organic result is holding your own name and nobody is taking meaningful share above it. Cut to a minimal presence, a low-bid Sponsored Brands defending the headline slot, and move the freed budget to non-branded terms where impression share actually changes what you sell.
Drop between 10% and 30%. Partial defense. Keep the campaign, but bid to the incremental number. If a third of the purchases were incremental, the campaign’s real ROAS is a third of what the report says, and the bid should be set against that figure, not against 10.8x.
Drop above 30%. Real conquest. Somebody is taking your traffic the moment you step aside. Fund the defense fully, screenshot your branded search results page to see exactly who is there, and check your branded impression rank. If you are not first on your own name, that is a finding on its own.
Why September is the only honest window
Three things change the return rate in Q4, and all three push it in the same direction.
Conquest rises. Competitors bid on rival brand names most aggressively in the ten weeks where a stolen click is worth most. The competitor who was not on your name in August may be on it in November.
Gift buyers are less loyal. A meaningful share of Q4 branded searches come from people who were told your brand name by someone else. They have no history with you, no loyalty to your organic result, and they will take the top card.
Prices are higher for everyone. Branded CPCs inflate with everything else, and if the recent FTC complaint about how Amazon prices sponsored clicks describes your account, the “cheap insurance” framing for branded spend deserves a check. Pull your CPC-to-bid ratio on your top branded terms while you are in the report.
Which means the share that returns organically in November will be lower than the share that returns in September. Run the holdout now and you have a baseline. Run it in November and you will be measuring a peak-specific number with nothing to compare it against, on the most expensive traffic of the year. Do not run a holdout inside the October 15 to January 14 window. The measurement week costs more than the information is worth.
Six things to do in the next two weeks
FAQ
Should I bid on my own brand name on Amazon at all? Almost always at some level, because the Sponsored Brands headline slot is either yours or a competitor’s. The question is never whether. It is how much, and the ROAS will not answer it. The holdout will.
Will pausing branded campaigns hurt my organic rank? Branded purchases keep happening organically during the pause, so velocity on those queries is largely preserved. Pausing non-branded campaigns is a different matter. If the risk worries you, use the SKU-split design rather than the time-based one.
What if the holdout shows a big drop but I cannot see any competitor on my name? Look at the mobile app, not desktop, and search from a logged-out session. Also check whether your organic result is actually first for your brand term. A weak organic position on your own name produces the same result as conquest, with a different fix.
How often should we re-run it? Twice a year. Once in September for the Q4 decision and once in Q1 when traffic normalises, because the competitive set on your name changes with every launch in your category.
Our branded campaign is our best ROAS. Why would we touch it? Because that sentence is the reason to test it. A number that good on a query set you already own is the single most likely place in the account for spend that is not doing anything.
If you’re looking for a team that manages every lever — creative, advertising, and operations — Velocity Sellers works with brands doing $100K+/month on Amazon. Contact us for a free account audit.
