Thirty to forty percent in 2021. Roughly eighty percent by 2024.
That is the share of Amazon Sponsored Products advertisers who paid their full maximum bid, according to the FTC complaint filed this week. Search Engine Journal covered the filing on Monday.
Read those two numbers next to each other and the entire case is visible without the legal language.
By 2024, four out of five winning bids on the largest retail ad platform in the world were paying the ceiling. Not the price required to beat the next advertiser. The ceiling.
What a soft reserve price actually does
The mechanism the FTC describes is a soft reserve price, introduced in 2019. It sets a minimum value for an ad placement that can raise the winning bid even when no competing advertiser required that level.
To understand why that matters, remember what a second-price auction promised.
You state your maximum. You pay only what is needed to beat the runner-up. The gap between the two is your margin, and it is the reason bid caps ever felt like a safety mechanism rather than a price.
A soft reserve quietly converts that into something much closer to a first-price auction, without the advertiser being told the rules changed.
The FTC estimates the practice generated more than $20 billion in additional advertising costs.
Amazon disputes the allegations, arguing that soft reserve prices are standard industry practice and that its auction model saved advertisers more than $8 billion between 2021 and 2025 through improved relevance calculations.
Both claims can be arithmetically true at the same time, which is precisely the problem. Relevance improvements and a hidden price floor are separate mechanisms, and netting them against each other in a press statement is not the same as disclosing either one.
Your benchmarks were measuring the wrong thing
Here is the operational reality for anyone who has run retail media in the last five years.
Every quarterly review I have sat in treats rising cost per click as a market signal. More competitors entered. The category got hot.
Q4 is expensive.
Demand went up, so price went up. That reading assumes the auction is a neutral instrument reporting the state of the market back to you.
If the FTC is right, a meaningful share of the CPC inflation on Amazon between 2021 and 2024 was not competitive pressure at all. It was a policy setting.
And the teams optimizing against it were doing careful, disciplined work on a number that was partly administrative.
I have watched agency teams rebuild bidding strategy three times in a single year chasing that curve. Nobody was wrong. Everybody was reading an instrument that was not reporting what they believed it reported.
Worse, the wrong conclusions compound. A team that reads a policy-driven CPC rise as competitive heat will usually respond by raising caps, narrowing targeting to defend efficiency, and cutting the tests that would have revealed the real cause.
This is the same structural issue as the AI Max black box problem I wrote about last week, only with a decade of history behind it. When the platform sets the price, runs the auction, reports the result and defines the metric, your optimization loop has no independent reference point.
It gets sharper as paid search is rebuilt for agents, because an agent bidding into an opaque auction on your behalf removes the last human who might have noticed the pattern.
Build measurement that survives a dishonest auction
The useful response here is not outrage. It is architecture.
Assume every closed auction you buy in may be adjusted in ways you cannot observe, and design your measurement so that assumption does not break you.
Three practical moves.
First, stop treating platform-reported CPC as a market indicator. It is a billing output. Track it, but do not build strategy narratives on top of its movement, because you cannot distinguish competitive pressure from a policy change.
Second, hold at least one measurement line the platform does not control. Incrementality tests, geo holdouts, matched-market comparisons, anything producing a number the seller did not calculate.
Most mid-market teams skip this because it costs sales volume for a few weeks. That short window is the only thing standing between you and a decade of unverifiable reporting.
Third, write your bid caps as commercial decisions rather than tactical ones. If a soft reserve can push you to your ceiling on most auctions, then your ceiling is not a safety mechanism. It is your actual price.
Set it at the number that still works for your margin when you pay it every single time, because on Amazon between 2021 and 2024 that is roughly what happened.
The broader point extends well past this case. We are moving into a period where more auctions are run by systems that are explicitly opaque, with AI bidding layers on both sides of the transaction.
The FTC needed internal documents and five years of data to make this argument about a mechanism introduced in 2019. Seven years, one regulator, one platform.
Nobody is going to file a case on your behalf about the next one.
The auction told you that you won. It never promised there was anyone else in the room.