On August 31, the FTC announced it had voted 2-0 to sue Amazon Ads, alleging it secretly rigged its ad auctions while telling advertisers the opposite.
If you’re not an Amazon Ads expert, it’s easy to pass this news by as just another petty dispute between a government body and large corporation.
This is anything but just another frivolous case: the suit alleges Amazon profited over 20 billion dollars from these surcharges over seven years of deception.
To understand exactly what the FTC is claiming and what truth there may or may not be, we need to unpack the nuances of Amazon Ads: why we’re talking about second-price auctions, how hidden surcharges can even exist, and what it has meant for sellers and brands who have spent over 200 billion dollars collectively on Amazon Ads in the last 7 years.
Some background on me: I’m Matt, co-founder and CEO of Laurence. We’re a tech company building quantitative models to autonomously run Amazon Ads for brands; I’ve spent the better part of my recent life deep in the weeds of Amazon Ads and how these auctions work under the hood, and I’m probably the most overqualified person on the internet to explain what’s going on here.
Without further ado, let’s dive straight in:
**
Table of Contents
- How Amazon Ads work: second-price auctions
- What the FTC says is actually happening
- What we’re seeing at Laurence
- What can brands to about this
How Amazon Ads work: second-price auctions
Amazon has two separate advertising businesses: their Sponsored Ads line and their DSP line. The lawsuit focuses on the Sponsored Ads line, which show consumers ads for purchasable products on Amazon.com primarily on keyword searches and product page views.
See that Sponsored tag under the bottom 5 product images? Those are advertised slots
(shoutout Fi, one of our amazing customers here in NYC!)
The key to understanding this lawsuit is understanding how these ads are shown and how brands pay for the ads: let’s focus on the bottom 5 products in the screenshot.
For each one of these products, the brand had to enter a bid ahead of time targeting the term dog tracking collar. Whenever dog tracking collar is searched, Amazon collects all the products that bid on the keyword and uses these bids to auction off the slots.
Typically, the highest bid wins the first slot, the second highest wins the second slot, and on and on. In reality Amazon ranks ads by bid times predicted click-through rate, but the pricing promise is the same: you pay just enough to beat the ad below you.
Brands only pay when an ad is clicked: this is called Pay-Per-Click and it’s why Amazon Ads are commonly referred to as “Amazon PPC”. This does mean an ad that gets shown but is never clicked is technically free.
Amazon claims that the highest bidder pays $0.01 per click more than the second highest bid, the second highest bidder pays $0.01 more than the third highest bid, and on and on. This is a second-price auction: advertisers call this cost-per-click the CPC.
What the FTC says is actually happening
“We don’t tell them about the surcharge, we let them assume [second-price auction].”
— Amazon employee, according to the FTC
The FTC’s core claim is simple: in 2019, Amazon stopped calculating CPCs the way we just walked through and never told anyone.
The allegation is that Amazon planted a “fake bidder” that drove your CPC up without changing anyone else’s bids. The fake bidder would never impact the ad placements or actual bids- it existed purely to increase CPCs advertisers paid to Amazon, thus increasing profits for Amazon.
The complaint provides evidence in form of a metric Amazon used internally: the “first-price rate”. This represents the share of clicks where the winner is charged their own full bid instead of a second price- essentially the share of clicks where the Amazon Ads auction behaves like a first-price auction rather than a second-price auction.
Amazon’s own reports put that number at 4% in late 2020. Then, per the complaint:
- 2021: between 30% and 40% of clicks
- 2022: 70%
- 2024: 79.1%, per an internal Amazon operating plan
First-price auctions aren’t inherently illegal: Google publicly switched to first-price auctions and Amazon’s other ads business, their DSP, is explicitly a first-price auction. The FTC is suing because it says Amazon’s Sponsored Ads auctions were first-price while Amazon disclosed them as second-price.
These are still just allegations, and Amazon will tell its side in court. But the complaint quotes Amazon’s own documents so extensively that the argument is less “trust the FTC” and more “read Amazon’s internal memos”.
What we’re seeing at Laurence
We’re incredibly well-positioned at Laurence to make sense of this news. We control bids for our customers and rigorously track CPCs as a core feature to our various statistical models.
We built our company on the fundamental thesis that advertising is a trading problem: advertisers on Amazon need to quantify value per click, model out the auctions they participate in, then maximize margin-adjusted bid-ask spreads in the form of total net profits over time. Just like how traders quantify value per asset, model out the public markets, then maximize PnL over time.
As a result, our approach to Amazon Sponsored Ads has always been quantitative. My cofounder Leo was a researcher at a quant hedge fund, and he’s spent at least half a year trying to build models and algorithms to maximize profits for our customers assuming the Amazon Ads auctions were second-price. All of these projects have not improved on our auction-naive approaches, and our conclusion internally was to stop assuming these auctions were second-price.
This is not to say we think the auctions are not second-price: this is something we can’t know for sure as a third party. We as a company have the utmost hope that we just made a series of mistakes and that the auctions are in fact second-price. We could have started our company building for any platform - Meta, Google, TikTok, even ChatGPT - but we started with Amazon because we believed in the platform’s durability and that we could provide the most immediate value to both brands and consumers on Amazon. I admire Amazon as a company like no other, and I’ll be the first person to support an evidence-driven Amazon rebuttal of this lawsuit.
However, the data we have is important to share because we’re the only player in this market with both access to the raw data and the technical expertise to break it down.
Across all our clicks, we found that we paid a CPC equal to our bid over 50% of the time.
This is interesting on two levels:
- Most of the first-price behavior is concentrated away from Top of Search This makes sense, since you can only show up for Top of Search if you have the absolute highest bid This is actually supporting evidence for Amazon that the auctions are second-price
- Even a 37% concentration of CPCs exactly at our bid was too high for us to do any meaningful math on top of the second-price auction assumption
We tried to build many different models on top of the second-price assumption: we wanted to actually model what each of our competitors’ exact bids were by using our CPC - $0.01 to determine the next highest bidder, but we just found that we our CPC equalled our bid too often for this to provide any meaningful signal. And any strategy that assumes there is non-zero gap between your bid and the next highest bidder breaks when we can’t actually measure the gap because of potentially inflated CPCs.
Ironically, the only bidding mechanism that made economic sense when tested was bid shading - deliberately bidding below a click’s value - which is exactly the optimal strategy when participating in a first-price auction.
So when this news dropped, it felt like the missing piece of the puzzle that explained half a year of dead ends.
What can brands do about this
Most of the problems that come from a first-price vs second-price ad auction world are solved through disciplined bidding. We realized early on that prioritizing robust value-per-click models based on
- each product’s true contribution margins
- organic rank and search volume per exact search term
- and conversion rate models for every placement
was far higher leverage than trying to solve the microstructures of the auctions.
This has led us to build a couple of features into Laurence’s autonomous Amazon Ads product that help brands spend more effectively on Amazon:
- value per-placement so we don’t pathologically overpay on Rest of Search or Product Page clicks this turns into a “placement-normalized bidding” strategy where you adjust your base bid down to your worst-converting placement and then adjust modifiers up to cover the placement gaps rather than only adjusting modifiers up
- actually calculating contribution margins per product for our customers, factoring in Amazon fees, returns
- LTV calculations for repeat-purchase products
We’ve been able to build a strategy that grows profits after margin and ad spend for brands by 30-40% by focusing purely on the value of owning what placements on which of the thousands of relevant search terms for each product, then autonomously bidding against that value to accrue profitable revenue.
But beyond the complex statistical models we’ve built to solve these problems, there’s not much we can do to solve this. Unless you have a quantitative system to autonomously tune bids, modifiers, budgets, and everything else to do with Amazon Ads every hour, you’re stuck playing catch up to numbers even your agents can’t read.
~
In any case, I look forward to a swift resolution here. I’m a massive believer in Amazon and there’s a reason Laurence started building for Amazon sellers- the ecosystem is there and I think it’s the best consumer experience on the internet.
Hopefully this article helped shine some light into what this lawsuit is actually about and provided some useful context into what an experienced math and data team has practically built around the Amazon Ads auction mechanics!
If anyone wants to reach me, my email is matthew@laurence.com. I’m a total Amazon nerd and love talking about these deep intricacies of the Amazon Ads platform, so always ready to chat :)
