The Alternative Mechanism: Information-Bundling Position Auction
The Information-Bundling Position Auction (IBPA) is a proposed answer to a specific problem in digital advertising: advertisers value different audience and context segments differently, but ordinary scalar bidding forces those preferences into one number. The mechanism keeps the realized segment private to the publisher while eliciting a valuation vector from each advertiser, allowing readers to see exactly what changes in the auction, how that change affects allocation and payment, and which claims remain limited to the paper's model.
From a bundle menu to a revenue curve
Start with one advertiser and one available ad slot. There are two possible segments, A and B. The publisher privately observes which segment an impression belongs to. The advertiser would pay a great deal for A, less for B, or perhaps the reverse. The advertiser knows their own values, but before the impression arrives they do not observe which segment will occur.
A scalar-bid system offers two familiar choices. It can reveal the segment and ask for a bid for that state, or it can hide the segment and ask for one bid for the pooled opportunity. The first choice enables precise matching but can divide the market. The second keeps the opportunity pooled but makes the advertiser pay for uncertainty. Neither choice lets the publisher screen the advertiser using the full pattern of values while keeping the realized state private.
IBPA begins with a menu instead. One menu option can offer the advertiser segment A only: the advertiser receives the impression if A occurs and receives nothing if B occurs. Another can offer segment B only. A third can offer the bundle: the advertiser receives the impression regardless of whether A or B occurs. Each option has its own price. The advertiser selects the option that gives them the highest expected utility based on their full valuation vector.
This is mixed-bundle pricing. Separate segment offers and a pooled bundle are available at the same time. It is related to second-degree price discrimination because the advertiser chooses among packages designed for different preference patterns; the advertiser is not simply quoted a single price for a known type. A reserve price in an ordinary auction is one screening threshold. A mixed-bundle menu replaces that single threshold with several choices that can distinguish an advertiser who values one segment, another advertiser who values the other, and a generalist who values both.
The menu matters because it changes what the publisher can learn from the advertiser's report. Under scalar bidding, the publisher sees a number associated with a disclosed block or an average over hidden states. Under IBPA, the report contains the relative value of the possible segments. The publisher does not need to reveal the realized segment before using that information to choose an allocation and payment. This is the mechanism's central separation: rich internal information and rich advertiser reporting can coexist with coarse or null disclosure.
The paper's formal construction does not simply choose one menu by intuition. It builds a revenue curve for each advertiser. For a given ex-ante probability of receiving the impression, the curve records the maximum expected revenue that a single-advertiser mechanism can extract while staying within that allocation limit. “Ex-ante” means before the segment and the advertiser's valuation are realized.