Targeting Can Make an Auction More Precise—and Less Competitive
Targeting information is the focus here: how audience and context data change what an impression is worth, who remains a credible bidder, and what a resulting CPM can mean. The key judgment is not that more targeting is good or bad; it is that information can improve the match between ad and person while simultaneously changing the thickness of the competition that produces the price.
The same impression has different values to different advertisers
An impression is not a uniform object. Its value can depend on the audience seeing it, the content around it, the device, the moment, and the action an advertiser expects after the ad is shown. A retailer selling running shoes may value an impression from someone who recently browsed athletic products more than one from a person with no such signal. A brand seeking broad awareness may value both opportunities more similarly. Two advertisers can therefore look at the same placement and attach different values to it.
The auction-design paper models this as a set of audience or contextual segments. The publisher privately observes which segment an arriving impression belongs to. Advertisers know the distribution of segments but privately hold a value for each one. That private value can represent expected downstream outcomes such as a click, purchase, or conversion, not just a general opinion about the placement. The important asymmetry is horizontal: the impression is not simply higher or lower quality for everyone. It is a better or worse match for different advertisers.
A conventional scalar-bid auction asks each advertiser to compress that multidimensional view into one number for the information available at bidding time. If the publisher discloses the realized segment, an advertiser can submit a bid conditional on that segment. A buyer who values segment A highly and segment B weakly can bid more for A and less for B. This improves the buyer's ability to avoid an irrelevant impression, but it also means that advertisers who care only about A may no longer compete in auctions for B, and vice versa.
If the publisher discloses nothing about the realized segment, each advertiser submits one scalar bid based on an expected value over the possible segments. That keeps more bidders in the same contest, but it can weaken matching. An advertiser that would pay substantially for A may be unwilling to bid enough for a pooled opportunity because the same offer also includes a chance of receiving B. Partial disclosure sits between these extremes: the publisher reveals a group containing the realized segment, and the advertiser bids on the expected value of that group.
The difference between internal information and disclosed information matters. The publisher may know more than the advertiser sees and can potentially use that private knowledge in allocation or pricing. Disclosure is a separate choice about what buyers learn before they bid. Treating the two choices as identical hides the design problem. A publisher can have fine-grained internal information while showing advertisers only a coarser category—or reveal the realized category and force the auction to work with the narrower set of conditional bidders.
Why scalar bidding creates a market-thickness problem
Scalar bidding is attractive because it makes a complicated auction operationally simple. One number can be ranked, compared with a reserve, and connected to a familiar payment rule.