Picture the usual account review slide. The branded Sponsored Products campaign runs at an 8-14% ACOS while everything else sits at 30-50%. It is the campaign everyone points to, and on a true-incremental basis it is often one of the weakest uses of ad dollars on the account.

The problem is the metric, not the campaign. ACOS and cost per order, as Amazon reports them, do not ask “did this ad cause a sale?” They ask “did a sale happen after a click on this ad?” For branded keywords, meaning searches that already contain your brand name, those two questions have very different answers, because a large share of those shoppers were going to buy from you anyway. The ad did not win the sale. It got there first and took the credit.

Three kinds of keyword, and why they behave differently

Account reports usually split keywords into branded and non-branded, and that split hides a third group. Own-brand, competitor-brand and generic terms each have a different job, so the same ACOS means something different in each.

In most reports, “non-branded” is a mix of competitor and generic terms. This article is about the first row, but the other two matter for the comparison later: the campaigns that look worse on paper are usually the ones where the ad is the only way the shopper finds you.

What a branded keyword actually captures

Someone typing your brand name has usually seen your product elsewhere, remembers a past purchase, or was told to buy it. The search is navigation, not discovery, which is why branded terms convert far better than generic ones. Compare the conversion rate on your own brand terms with the rest of your search term report and the gap shows up in your own data.

That shopper also has free paths to your listing. In most cases your organic listing ranks at or near the top for your own name, and returning customers can reach your Store or past orders directly. A sponsored ad on that term is one more path competing with paths that already existed. Check this before relying on it: search your brand name the way a shopper would and look at what sits above your organic listing. A newer brand, or one whose name is also a common word, may not own that page yet, in which case the ad is doing more work than this article assumes.

Before you measure anything: separate the pools

Brand queries leak into non-branded campaigns. Auto and broad-match campaigns pick up searches for your own name, so some brand orders land in the non-branded column, which flatters those campaigns and means you may be paying twice for the same shopper. Filter your search term report for your brand string across every non-branded campaign, see how much spend and sales sit there, and add your brand terms as negative exact matches. Until you do, neither side of the comparison below is clean.

The core problem: reported metrics assume every order is incremental

Reported cost per order treats every attributed order as caused by the ad:

Reported cost per order = Ad spend ÷ Attributed orders

That is a reasonable approximation for a keyword nobody would reach you through otherwise. It is a poor one where the buyer already knew where they were going. The number that matters is the true cost per incremental order, which strips out the orders that would have happened anyway:

True cost per incremental order = Ad spend ÷ (Attributed orders × (1 − OCR))

OCR stands for Organic Capture Rate, our own shorthand rather than an industry-standard metric. It is the share of attributed orders that would have converted through a free channel (organic ranking, your Store, a direct link) if the ad did not exist. At OCR = 0%, true cost equals reported cost. At OCR = 90%, true cost is ten times the reported number, because nine in ten of the ad’s “wins” were sales you already had. The underlying idea is what advertisers elsewhere call incrementality or conversion lift, and it is measured the same way.

In practice you measure OCR from what disappears when the ad is switched off:

OCR = 1 − (Orders lost when the ad is off ÷ Attributed orders)

Two ways to estimate it

1. A holdout test. Pause brand targeting on a matched set of ASINs for 2-4 weeks while a comparable set keeps running, and compare total purchases on your brand queries for each set. Sponsored Products cannot be split by region, so the clean split is by ASIN rather than geography. With only a few ASINs, alternate on and off windows on the same products and use unpaused ASINs to control for seasonality. Compare total purchases, paid plus organic, not ad-attributed orders, which fall to zero by construction. If you are brand registered, the Search Query Performance report in Brand Analytics gives total purchases by query and ASIN.

Here is how the arithmetic works. Say your brand queries produce 1,000 total purchases in the period, 400 of them attributed to the ad. With the ad paused (and adjusted against the control group), total purchases fall to 940. Sixty orders were lost, so OCR = 1 − 60 ÷ 400 = 85%. Three hundred and forty of the 400 attributed orders came back through free paths.

2. The new-to-brand rate as a cheap proxy. If a shopper converting on your brand keyword is not new-to-brand, they have bought from you before and know how to find you:

OCR (proxy) ≈ 1 − NTB rate on branded keywords

Availability matters here. New-to-brand columns come standard in Sponsored Brands and Sponsored Display reports, but at the time of writing, term-level new-to-brand for Sponsored Products generally requires Amazon Marketing Cloud, which not every seller has. If you run Sponsored Products only and have no AMC access, skip this method and rely on the holdout and the bid step-down later in this piece.

Two limits on the proxy. It only works on branded keywords: a repeat buyer typing a generic query may still have needed the ad to find you, so the same arithmetic would badly overstate OCR there. And Amazon counts a buyer as new-to-brand if they have not purchased from the brand in the past 12 months, so lapsed customers show up as new, which makes the proxy conservative. Use it to decide which campaigns deserve a real holdout, not as your final number.

A worked example (illustrative figures)

An account runs one branded and one non-branded campaign at a $35 average order value. Both campaigns go through a 3-week holdout like the one above:

Side by side as reported, the branded campaign looks like the star: $5 per order against $16. Adjusted for what the ads actually caused, the ranking flips: $33.33 per incremental order for branded against $17.39 for non-branded. The branded campaign also runs a true ACOS of 95%, meaning nearly every dollar of sales it truly adds is spent on the ad. The non-branded campaign loses only 8% of its orders to cannibalization because almost none of those shoppers had another way to find you. The campaign with the worse-looking scoreboard is doing the real work.

The new-to-brand row shows the proxy at work. 9% new-to-brand on the branded campaign implies an OCR of about 91%, against the 85% the holdout measured: close enough to triage. For the non-branded campaign, 71% new-to-brand would imply an OCR of 29% against the 8% measured, which is exactly why the proxy should not be used outside branded terms.

When this does not apply

The argument depends on the shopper having a free path to you. It is weakest for a new brand with little name recognition, where your organic listing may not yet rank first for your own name. It is weakest in categories where competitors aggressively bid on your brand, because the ad is then protecting the sale rather than duplicating it. And it is a distraction if branded spend is a small share of the account, in which case the same hour is better spent on the non-branded tiers. A holdout will tell you which of these situations you are in, which is the reason to run one before cutting anything.

The second leak: overpaying for the defense itself

Incrementality is one way branded spend gets wasted. The other has nothing to do with it: even spend that is worth making is often priced too high. Brand bids tend to be set high once, out of caution, and never revisited, and top-of-search placement modifiers stacked on a high base bid can push the real CPC well above what the auction requires to hold position.

The test is a bid step-down. Lower the max bid on a branded keyword in steps of 20-25% every week or two, and watch top-of-search impression share and orders at each step. The number that tells you when to stop is the marginal cost of the orders you give up:

Cost per order given up = Spend saved ÷ Orders lost, step to step

Here is the same branded campaign from the example, with 2,000 clicks a month and a 20% conversion rate, stepped down. Assume a 30% contribution margin, so each order is worth $10.50 to you.

The first step saves $280 a month and loses nothing. The second saves $307 and loses two orders. By step 3 the campaign has saved $875 a month and given up 12 attributed orders, $72.92 of spend per order, on a product with a $35 order value. Step 4 is where defense starts to matter: top-of-search share falls from 93% to 81% and the cost of the orders given up drops to $6.77, below the $10.50 an order is worth. Hold at step 3. That is the defensive floor for this keyword, and it costs about 44% less than the starting bid. These are the same first-difference mechanics as a marginal CAC curve, applied to a single keyword.

Because 85% of those attributed orders would likely have arrived anyway, valuing each at its full $10.50 makes this a conservative test. A step-down is also cheaper and faster than a holdout, since it needs no pause and no control group. Competitor bids change, so re-run it quarterly.

The decision rule

Treat branded spend as two layers. The first layer is the defensive floor: the bid you reach by stepping down for as long as each step saves more than the orders it costs you are worth, which is what the step-down finds. Hold that spend, because it protects the sale. The second layer is everything above the floor, which buys neither defense nor incremental sales and should be cut.

What you do with the freed budget needs one more check. Moving it dollar for dollar into non-branded keywords assumes non-branded demand can absorb it at its average cost, and it cannot: each extra tier costs more than the one before. Move budget only while the true cost per incremental order on branded spend exceeds the marginal cost per order of your next non-branded tier, and stop when they meet. If your branded keywords show a new-to-brand rate under roughly 20% and meaningful spend, test those first, since that is where the gap is most likely to be large.

Caveats. One holdout is noisy: seasonality, promotions or a competitor stockout in the test window can all move the control number, so treat a single test as a strong signal and re-test before a large permanent shift. Pausing too much brand spend at once can hand the top slot to a competitor for weeks. Trim in steps and watch total brand-query purchases, not campaign-reported orders.

How to start this week

• Filter your search term report for your brand string in every non-branded campaign, and add those terms as negative exact matches.

• Search your brand name as a shopper would, then pull top-of-search impression share on your biggest brand keywords to see whether anyone is actually competing for them.

• Start a bid step-down on your top branded keyword: lower the max bid 20-25%, wait a week, and log spend, orders and impression share at each step.

• If branded spend is large, run a holdout on a matched set of ASINs and compare total brand-query purchases. If you have AMC, pull new-to-brand on branded terms to pick which campaign to test first.

• Move freed budget into non-branded tiers only while their marginal cost per order stays below your branded true cost, and recheck total account orders after 60 days.

If you would like a second set of eyes on which branded keywords are earning their spend and which are riding on demand you already owned, we run that audit alongside the account.