Advertisers can participate in repeated auctions without learning how their charges are set when platforms report only win rates and average costs, according to October research by University of Toronto economist Joshua Gans.
His National Bureau of Economic Research working paper examines a pricing mechanism called a soft reserve. The model separates an advertiser’s ability to measure campaign performance from its ability to identify the auction rule behind the bill.
The subject has become commercially relevant following an August lawsuit by the US Federal Trade Commission and 22 states against Amazon. The regulators allege that Amazon misled advertisers about auction pricing. The case is pending, and Amazon disputes the allegations.
Gans models the economic arguments conditional on the pricing mechanism being present. His research neither establishes what Amazon did nor decides whether its conduct was unlawful.
A winning bid and the price paid are different amounts
In a basic second-price auction, the highest bidder wins but pays the amount needed to beat the next competing bid, subject to the auction’s rules. Bidding more can improve the chance of winning without automatically raising the price to the full amount offered.
Real advertising auctions can also adjust rankings for relevance and other factors. A bid is consequently not the only determinant of which advertisement appears.
In the paper’s model, a soft reserve can raise the winner’s charge above the competitive threshold, up to specified limits, including the advertiser’s own bid. The competitive threshold is the amount needed to win given the competing advertisements.
A hypothetical advertiser might bid $10 for a placement it could otherwise have won for $6. The modeled pricing rule can raise the charge above $6 without exceeding $10. Paying no more than the submitted bid does not establish that the advertiser received the pricing process it expected.
The winner usually cannot observe the competing bids. A single charge, considered on its own, can look consistent with several different explanations.
The reporting format affects what advertisers learn
Gans compares two reporting arrangements. One provides win rates and expected charges. The other also identifies individual wins in which the charge equals the advertiser’s bid.
Under aggregate reporting, the model allows the soft-reserve mechanism to look like a second-price auction at any fixed bid. An advertiser that fully trusts the second-price description and bids its valuation can remain confident in that description without learning the underlying rule.
An advertiser that suspects a different rule can experiment with lower bids. Learning may still be slow when the initial suspicion is weak and the advertiser chooses bids to maximize current profit.
Identifying individual charges equal to bids supplies different information. Under the paper’s assumptions, including continuously distributed competition without exact ties, that information can expose the mechanism much faster.
Those assumptions limit the result. In practice, exact ties, changing competition, relevance adjustments and reporting conventions need examination before a particular payment is treated as proof of a surcharge. The paper also distinguishes expected charges from every possible statistic an advertiser might derive from a finite campaign.
Campaign returns cannot answer every pricing question
In its response to the FTC lawsuit, Amazon says advertisers adjust bids using observed performance and disputes that they were misled. It argues that its emphasis on ad relevance benefits advertisers and shoppers.
The model asks a separate question: whether the feedback available makes a particular pricing mechanism identifiable. An advertiser can monitor sales and decide that a campaign is worth continuing without being able to reconstruct the counterfactual price under another auction rule.
Our guide to click-through rates and marketing decisions explains why clicks alone cannot establish commercial returns. The auction research adds another reporting issue: an average cost can describe a campaign while concealing information about how individual charges arise.
The paper is theoretical, has not been peer-reviewed and does not estimate a universal overpayment rate. Gans discloses earlier consulting work for Amazon on unrelated issues.
For advertising buyers comparing platforms or agency reports, the analysis gives a reason to ask what transaction-level pricing information is retained, which auction rules apply and whether reported averages can be reconciled to individual charges. Campaign performance and pricing transparency require different evidence.
