The method
Placement architecture on Amazon Sponsored Products
Amazon gives you one placement lever per campaign. Here is what that costs, and what we do about it.
The problem
Amazon lets you adjust bids by placement, from 0% up to 900%. Top of search, rest of search, product pages, and since early 2025 Amazon Business as well.
Every one of those adjustments works at campaign level. It applies to every keyword and every target underneath it, all at once.
That is fine until the targets inside a campaign behave differently from each other, which is most of the time. A keyword can convert well when a shopper sees it on a product detail page and lose money when the same keyword surfaces in search results. Inside a single campaign there is no adjustment that treats those two situations differently. Raise top of search and you fund the losers alongside the winners. Lower it and you starve the winners along with the losers.
This is not a reporting problem. It is a control problem, and the two have different solutions.
Why better data does not fix it
The obvious response is that you just need better reporting. That response is half right, and the half that is wrong is the important half.
Amazon's standard console reports placement performance at campaign level only, aggregated across every keyword and target beneath it. The standard v3 reporting API is the same. Placement is available in the campaign report and nowhere else. The targeting report and the search term report carry no placement dimension at all.
So within the tooling most advertisers use, the target-by-placement view does not exist.
But two Amazon products do expose it. Amazon Marketing Stream delivers placement and keyword in the same record. Amazon Marketing Cloud carries placement type alongside targeting and search term in the same table. A team with the engineering to consume either one can see this problem clearly.
Seeing it does not give you a lever. Amazon's placement control is still campaign-level no matter how good your reporting is. You can know that a keyword earns its money on product pages and loses it in search, and have nothing inside that campaign to act on.
That is the gap this structure closes.
What we do instead
We run search-focused and product-page-focused campaigns separately. Each one carries its own budget, its own bids, and its own read.
The mechanism is a suppressed base bid combined with a high multiplier on the intended placement. The base bid sits low enough that unintended placements rarely win, and the multiplier funds the placement the campaign exists to serve.
That only holds on the right bidding strategy. We run bid-down-only or fixed, never dynamic up-and-down, and the reason is mechanical. With up-and-down enabled, Amazon can raise a bid by as much as 100% for top of search, and that increase compounds on top of the placement multiplier. A suppressed placement can be bid back up by the platform. On down-only or fixed, the low base bid is a ceiling and that cannot happen.
Anyone who copies the structure onto up-and-down bidding will watch the suppression fail and conclude the structure does not work.
What we will not claim
Amazon does not offer placement exclusion. A 0% placement adjustment removes the bid increase, it does not stop delivery, so this is deliberate weighting rather than a partition. Some delivery reaches the placements we steer away from, and we measured how much.
Measured across $24 million in Sponsored Products spend, in periods excluding promotional events, tentpole days and seasonal peaks, because those distort placement distribution.
No campaign age filter was applied.
Read each pair together. In both directions the leaked delivery earns more sales per dollar than the campaign it sits inside, roughly three and a half times the campaign average in product-page campaigns and twice the average in search campaigns. The leak is not waste. It is the cheapest inventory in the campaign, because a suppressed bid only wins the auctions nobody else wanted at that price.
The two directions are not worth the same, either. A product-page campaign leaking into search captures net-new demand. A search campaign leaking onto product pages may re-buy traffic already in the funnel. The two figures look alike and are not equivalent.
That efficiency does not scale. The leak is cheap because it is small and self-selects for auctions nobody else wanted at that price. Push volume into it and the advantage disappears.
We also do not split everything. Below a certain click volume, dividing a target's traffic across two campaigns means each side accumulates signal too slowly to act on, and Amazon data has a shelf life. Seasonality, price changes, competitive entry and inventory state all shift underneath a slow-accumulating sample, so a split that takes six months to reach significance can return a stale answer rather than a slow one. We set the threshold per account against actual click volume and conversion rate, and below it we keep one campaign and manage with multipliers.
The objections, answered
Every argument below is one we have had put to us, and each one has something to it. The concession comes first because the concession is what makes the answer worth reading.
A new campaign loses its learning history
It is, and it has a price. Velocity Sellers puts a campaign clone or split at two to four weeks of elevated ACoS on the migrated spend, typically 20% to 40% above baseline.
We do not migrate wholesale. We review the history, leave the established campaign on whichever placement its sales already skew toward so it keeps that history, and build the counterpart for the minority placement, which was under-served inside the aggregated campaign anyway.
The recalibration cost applies to that slice rather than the account. New accounts launch split and never pay it at all.
Splitting starves each campaign of data
Published thresholds put reliable keyword-level judgment at roughly 15 to 20 clicks minimum at a 10% conversion rate. Halve the traffic and you halve the rate at which a target reaches that.
Above a volume threshold this is a one-time transition cost against a permanent benefit, because aggregated placement data taxes every decision for the life of the account.
Below the threshold the argument does not hold, and that is exactly why the gate exists rather than a blanket rule.
Two budgets cannot help each other
They cannot. A budget in one campaign is unavailable to the other.
A budget-capped account forfeits impression share whether it runs one campaign across placements or separate campaigns per placement. The objection does not distinguish between the two structures, so it is not an argument against this one.
The residue is operational rather than economic. Splitting doubles the number of budget lines where a capping error can happen.
Doubled negation maintenance
The maintenance surface doubles. Two campaigns means two negation lists to keep in step.
Keyword negation and ASIN negation carry different thresholds, and both run at daily or better cadence with tooling support.
Your own campaigns will compete against each other
Deduplication does happen, and it is worth understanding rather than waving away.
Amazon does not let one advertiser's ads compete in the same auction, and generally enters only the higher bid, so average cost per click is not inflated by the split.
The second-order effect is worth stating honestly. Where both campaigns are eligible for the same search impression, one suppresses the other, which can slightly distort the placement read the split was built to clean. That is worth monitoring rather than worth avoiding.
Marketing Cloud already shows this
Completely. Amazon Marketing Cloud carries placement alongside targeting and search term in the same table, and Marketing Stream delivers placement and keyword in the same record. The data exists.
The question the objection does not answer is what lever you pull once you have it. Amazon's placement control remains campaign-level regardless of how the data reaches you. Observability is not controllability.
Where this does not apply
This is the wrong structure for a low-volume catalog where individual targets do not clear a decision threshold on their own. It is the wrong structure for an advertiser running dynamic up-and-down bidding who does not intend to change that. And it is unnecessary for a team already consuming Amazon Marketing Stream or Marketing Cloud who has built the tooling to act on placement-level reads inside a single campaign, which is a real if uncommon setup.
If you want to know how your own placement split is performing before you change anything, that is the first thing we look at.
Sources
Amazon documentation
- Amazon Ads, rest of search bid adjustment announcement, January 2024
- Amazon Ads, dynamic bidding up and down guide, for the up-to-100% top-of-search increase
- Amazon Ads targeting documentation, for bidirectional serving
- amzn/ads-advanced-tools-docs on GitHub, for v3 reporting dimensionality
- Amazon Ads, introducing Amazon Marketing Stream, which states that Stream enables the combination of targeting expression and ad placement
Third party
- Velocity Sellers, August 2026, restructure recalibration cost
- Ad Badger, bidirectional serving, and the figure that over 67% of clicks go to first-row search results
- Tinuiti, Amazon Marketing Stream placement joined to keyword
- SellerMetrics and Adbrew, click thresholds for reliable decisions
- Brandwoven, an independent version of the placement separation argument