Soft Reserves and Amazon Ad Costs: What the FTC Case Means for Sellers

This note explains how soft reserves work and how bid shading can help reduce advertising costs.

FoxtrotAI · October 2026 · 9 min readOriginal research by FoxtrotAI, based on public filings and disclosures. The analysis and opinions are our own.

What the FTC alleges, and what Amazon says

The FTC complaint says Amazon long described its auctions as generalized second-price: the winner pays the calculated second price, the CPC that would just displace the runner-up. The FTC alleges that Amazon quietly added a soft reserve after each auction and charged the higher of the auction price or its own computed price for the click, up to the advertiser's bid. It says a pricing method using Amazon's estimate of the sales a click will generate, called "eOPS," was tested in 2021 and used for almost all Sponsored Products search ads from 2022.

Amazon's ad revenue growth shows an unusual shift in that same timeframe. In Q2 2021, Amazon's advertising revenue growth jumped to about 6x its paid unit growth, and it kept climbing until Q2 2022, far above the long-term median of about 2.1x.¹

Each multiple divides one growth rate by another, so it can rise because ad revenue grew faster or because unit sales grew slower. The chart below plots the two rates separately.¹

Sponsored Products ads look similar to organic product listings and are embedded within search and product detail pages. Since units and sponsored ad revenue come from the same shopping sessions, we can expect their growth rates to move together:

Shopper traffic, clicks, and conversion drive both. Ad impressions can grow faster than organic impressions as ad density increases, driving faster ad revenue growth. Indeed, in most years, Amazon's ad revenue grew about twice as fast as units: in 2019, and again in 2024 and 2025. However, an increase in ad slots naturally exerts downward pressure on the average CPC (as lower-ranked ads get selected), so there is a limit to how much ad revenue can grow faster than units purely from ad density shifts.¹

In Q2 2021, Amazon's ad revenue growth and unit growth diverged significantly. Ad revenue grew 88 percent while units grew 15 percent, and ad revenue kept growing several times faster than units through Q2 2022. Higher cost per click from soft reserves, as the FTC alleges, could be a key reason. This analysis is necessarily imprecise, and there are other puts and takes. In 2020, ad revenue grew more slowly than usual relative to units, so some of the 2021 divergence may be ad revenue catching up. On the other hand, as the room to add ad placements shrinks, it gets harder for ad revenue to keep growing twice as fast as units. In 2024 and 2025, the median multiple was 1.8x, just under 2x, even though Prime Video ads, which do not depend on unit sales, started in 2024.¹

Amazon denies misleading advertisers, but its response confirmed that it uses soft and hard reserves. This note takes no position on the FTC case. Instead, it explains how reserves work and how bid shading can reduce the impact of soft reserves, and it explores the potential size of the opportunity.

How hard and soft reserves set your price

Each auction has a hard reserve, the minimum a bid must clear to enter, and a soft reserve, which Amazon describes as a real-time value that seeks to reflect what a placement is actually worth. A winning bid that clears both reserves pays the soft reserve. A winning bid that clears only the hard reserve pays the bid itself. The explorer below follows Amazon's description, with the reserves fixed at $0.50 (hard) and $1.60 (soft) and the second price at $0.80.² Move your bid to see what you pay. Switch the soft reserve off to revert to a plain second-price auction and compare.

The FTC complaint says Sponsored Products advertisers paid their full bid on close to 80 percent of clicks in 2024. In the explorer, any winning bid below the soft reserve pays in full. So by 2024, soft reserves were likely above the winning bid on most clicks, and most winners paid exactly what they bid.

The lesson for bidding: between the second price and the soft reserve, every extra cent of bid is a cent paid. The ideal bid is just above the second price, high enough to keep the placement and low enough to pay little more than the auction requires.

That is what bid shading does. Bid shading means bidding less than a click is worth, as close as possible to the price needed to win. Once soft reserves make winners pay their own bid on a large share of clicks, bidding full value means paying full value. Bid shading allows advertisers to keep the difference. So how large is that difference?

Sizing the bid shading savings opportunity: an illustrative analysis

The ad revenue growth chart above suggests Amazon's ad revenue jumped far ahead of unit growth. As a hypothetical analysis, we built a counterfactual scenario: what Amazon's ad revenue would have been from Q2 2021, when the jump began, if it had kept growing at 2.1x unit growth.³ The difference between that scenario and Amazon's reported ad revenue is an illustrative upper bound on what soft reserve pricing could have added, and so on the room bid shading has to work in.

The 2.1x peg is the historical median ratio of ad revenue growth to unit growth. You can adjust this peg in the interactive below to see different scenarios.

At 2.1x, counterfactual ad revenue from Q2 2021 through Q4 2025 totals $173.8 billion, against $234.3 billion reported. The difference is $60.5 billion, or 25.8% of reported revenue. If soft reserves explained all of it, advertisers could have bought the same clicks for up to 25.8% less. Other growth drivers could explain part of the gap, and the figures cover all of Amazon's advertising, so the percentage does not apply directly to any one seller.

Try your own assumption below. The chart begins a year before the peg starts so you can see where the two paths diverge. Drag the multiple to adjust the scenario.

That 25.8 percent is a hypothetical ceiling for the chosen peg. A 3.0x peg would cut the gap to $21.3 billion, or 9.1 percent, in the interactive above. And even if this estimate is in the ballpark, sellers cannot recoup all of it. The gap will vary by seller, there are limitations in the data available to do effective bid shading, and a bid shaded too far loses placements.⁴ Nonetheless, the analysis suggests that bid shading is an opportunity worth exploring for sellers.

How bid shading works

Most sellers decide what they are willing to pay for a click from a target ACoS. For a $25 product with a 10 percent conversion rate, a 20 percent target ACoS supports a cost per click of about $0.50. Bid shading works on top of that number: it asks whether bidding less than $0.50 earns more profit overall.

Shading is a trade-off. A lower bid keeps more profit on each click you win but wins fewer auctions. A higher bid wins more often but keeps less, and a bid at the click's full value keeps nothing. The best bid is the one that maximizes expected profit per auction: the chance of winning times the profit on each click won.

The illustration below lets you find it. For simplicity, it assumes you pay your bid on every click you win. On Sponsored Products that was true of close to 80 percent of clicks in 2024, per the complaint. The logic is the same, but the real savings are somewhat smaller than the illustration shows.

In this example, bidding the full $2.00 wins 92 percent of the time and keeps nothing, because every click costs exactly what it is worth. Bidding about $1.02 wins 62 percent of the time and keeps about $0.98 on every click it wins. Real auctions are messier, but the logic holds wherever you often pay what you bid.

A proven practice in other ad markets

Bid shading is an established practice in the display ads market. Display ads are the banner and box ads you see on websites and apps, such as the banners on a news site like The New York Times. Unlike Sponsored Products, which charge per click, display ads are usually priced by CPM, a cost per thousand impressions, so advertisers pay each time an ad is shown. Much of this inventory is sold automatically. Each time a page loads, its ad space is auctioned on an ad exchange in a fraction of a second. Advertisers rarely bid on each impression themselves. They use a demand-side platform (DSP), software that values each impression and bids on their behalf across many exchanges.

Display advertising moved from second-price to first-price auctions starting in 2017. Major ad exchanges switched first, and Google Ad Manager, the largest, followed. Buyers responded with bid shading, which became a standard feature of DSPs.

• Amazon DSPAmazon describes bid shading as a standard service it provides to all of its DSP advertising customers at no cost on third-party supply. Its technical whitepaper works through an impression worth $5.00 and finds an optimal bid of $3.39. Amazon Ads whitepaper
• Pan et al., AdKDD 2020 (Verizon Media DSP)Shading reduced price per impression to about 55 percent of the unshaded cost, and the authors' win-rate method captured 7 percent more advertiser profit than a common baseline approach. arXiv:2009.09259
• Zhou et al., KDD 2021A deep learning shading model raised advertiser ROI by 2.4 percent for click-based campaigns in live A/B tests. arXiv:2107.06650

These results come from display auctions with bid-level feedback. They are not reflective of what bid shading can achieve with Sponsored Products.

Why shading on Sponsored Products is harder but still worth doing

If shading is standard in DSPs, why is every seller not doing it on Sponsored Products? It is harder. A DSP sees every bid request it enters and whether it won. Sponsored Products advertisers work with much less:

1. No auction-level data. Amazon reports keyword-level totals: impressions, clicks, and average cost per click. You never see individual auctions, competing bids, or the price that would have won, so win rates and costs have to be estimated from those totals and from deliberate bid changes.
2. Placements cannot be bid separately. A single search can show many ad placements, each with its own auction, but you set one bid per keyword or target. Amazon groups placements into three buckets (top of search, rest of search, and product pages) and lets you raise the bid for each bucket by a percentage. You cannot shade the bid for an individual placement or position.

Even with these limits, shading is worth doing. When winners pay their own bid most of the time, a bid closer to the second price keeps more of every ad dollar. Doing it well means estimating win rates and costs for every keyword and placement bucket, and re-estimating them as competition, seasonality, and reserves shift. Across hundreds of keywords, that takes automation.

Sellers with Amazon Ads API access can build their own bid shading strategy, or they can work with their agency to add it to their bidding. FoxtrotAI also includes bid shading in its bidding engine.

← All research notes