A single keyword’s bid doesn’t determine one outcome — it determines a starting point that gets adjusted depending on where the ad actually appears. Amazon groups placements into three types, and each one carries a meaningfully different level of purchase intent.
Top of Search (ToS)
The first 1–3 ad positions right above organic search results — the most visible, most competitive real estate on the page, and the placement with the highest purchase intent. As a general benchmark, a healthy ToS ACoS target is often under 30%.
When to raise the modifier: ToS is converting below your ACoS target and you want more of that high-intent volume. When to cut it: ToS ACoS is running well above target — you’re paying a premium for the position without the conversion rate to justify it.
Rest of Search (RoS)
Page two and beyond of search results. Lower visibility, lower purchase intent, and generally the lowest-quality traffic of the three placement types — a target ACoS under 25% is a reasonable benchmark here, tighter than ToS, since the traffic itself converts less reliably.
When RoS ACoS runs very high, the usual move is to zero out or significantly cut the modifier rather than try to optimise it — this placement rarely rewards extra spend the way ToS does.
Product Pages (PP)
Ads shown on other products’ detail pages — mid-level purchase intent, since the shopper is actively comparing options rather than just starting a search. A reasonable ACoS benchmark here sits a little higher, often under 35%, reflecting that intent level. This is also the placement PAT/ASIN-targeting campaigns rely on almost entirely.
Why one blended bid isn’t enough
Without placement modifiers, a single bid gets applied uniformly across three positions with three different intent levels and three different realistic ACoS ranges — effectively overpaying in one placement to compensate for underpaying in another. Reviewing placement performance separately, and adjusting the modifier for each rather than the base bid alone, is how the same keyword ends up appropriately priced in all three places instead of averaged across them.
One practical note: after changing a placement modifier, give it at least seven days before drawing conclusions — Amazon’s auction dynamics take time to settle, and a few days of data after a change is rarely enough to judge it fairly.