Advertisers tuned seven years of Amazon bids to a rule the FTC says applied one time in five

The bid field is the only real control in a Sponsored Products campaign. Under the rule Amazon described, that field is a ceiling: you enter the most a click is worth to you, and the auction returns a price set by the runner-up plus a cent ^[3]. That is why the format spread in the first place: when somebody else's bid sets your price, honest bidding is the rational move ^[6]. Under the rule the FTC alleges, though, the number you typed is the number you paid ^[4], and on the screen the two situations look identical.

None of that breaks a ROAS report. Spend was real, revenue was real, and the ratio held whatever the auction was doing underneath, which is how a discrepancy sits in an account for seven years without reading as an anomaly ^[2]. What stops working is the inference that runs backwards from the report to the bid. If you raised a target CPC because reported CPC kept landing under your ceiling, you were reading a number you had set yourself.

Amazon's answer is that advertisers optimise against results, not against descriptions of mechanics ^[7]. The counter, as TNW puts it, is that performance optimisation only works when the translation from bid to price is stable, so rules that differ from the published ones make that translation less predictable rather than irrelevant ^[8]. Both can be true at once for a large advertiser running weekly bid experiments. Neither does much for the seller who set a bid per ASIN in 2019 and left it.

The two figures in circulation do not net out. Amazon says accounting for ad relevance saved advertisers more than $8bn between 2021 and 2025 ^[9]. Its advertising business took $68.6bn in 2025 alone, third behind Google and Meta ^[10]. Divide one by the other and five years of claimed savings comes to roughly 11.7% of a single year of revenue ^[11], and it answers a question about which ad wins rather than what the winner pays. By the FTC's own count, the pricing behaved as advertised in about one auction in five ^[12].

The diagnostic is cheap enough to run before any of this resolves. Take matched campaigns, cut bids 10 to 15%, change nothing else, and watch CPC against impression share. If CPC falls roughly in step with the bid, you are paying your own number and the ceiling was never a ceiling. If CPC holds while impressions thin out, someone else's bid is setting your price.

Then sort accounts on two questions. Did you set bids from a value ceiling or from an observed CPC target, and have you ever tested bids downward while holding volume. The quadrant with value ceilings and no downward test is where unmeasured overpay lives, and it is populated mostly by mid-size sellers who depend on Amazon for distribution and have the least room to stop buying sponsored listings ^[13].