Puffin

From Bitnami MediaWiki
Revision as of 15:50, 17 November 2021 by Jkoran (talk | contribs) (→‎Impact)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Cafemedia’s Personal User Floor For Impression Negotiations (Puffin) aims to improve publisher inventory monetization by setting a per-browser, per-ad type price floor based on historical auction data for each placement by ad category for each publisher.[1] The proposal aims to increase value for content relative to interest-based advertising, by increasing the price floor solely for interest-based ads. This delta in pricing suggests more context-only ads would win auctions relative to today. The notion is that increasing revenues for high-quality content producing sites encourages greater production of such content "in the interests of the user and of the ad-supported sites whose content they prefer."[2]

Puffin adopts Google’s Privacy Sandbox goal of disintermediating publishers from the marketers that fund their digital properties.

Overview

Puffin builds on the trusted-server model of required by Turtledove/Fledge. Unlike Fledge, Puffin allows real-time, per publisher per ad unit per browser price floors based purely on prior auctions that relied solely on current context.

"The interest-based ad wins (and is displayed to the user) if it can outbid the contextual ad. A PUFFIN is a persistent per-browser floor price that the interest-group ad bid must also exceed in order to win the auction. Each PUFFIN is calculated based on an exponentially weighted rolling average of winning contextual ad bids. Losing bids and interest-group bids never affect PUFFIN."[3]

The price floors are stored in the browser for each ad unit type and placement per publisher. "The browser maintains one PUFFIN for each ad unit type that can be detected automatically. For example, a video ad and large leaderboard ad would each have a different PUFFIN from a small, below-the-fold ad."[4]

The price floors are also calculated per ad category, such that auto ads may have a different floor than travel ads. "With PUFFIN, in order for the interest-based ad to win the in-browser auction, it must beat not only the highest-bidding contextual ad, but also the PUFFIN for ads of the same category that have previously appeared in the same browser."[5]

For example, if a given ad slot usually monetizes via context at $1.00, then an interest-base ad must always exceed $1.00 even if all context bids for a given auction were only $0.50. In the case that an interest-based ad were $0.90, Puffin would have the browser choose the lower monetizing contextual-only ad. Note: as this example illustrates this will reduce yield if publisher inventory is forced to monetize for the lower priced ad.

Unlike other interest-based advertising proposals (e.g., Pauraque), Puffin does not expose any of information to auction participants. "PUFFIN is confidential and per-browser, and is not available to any script."[6]

Impact

Unlike other Privacy Sandbox proposals that nominally allude to improving people's privacy, Puffin is designed to ask the browser to steer money towards certain sites based on analyzing people's online activity. People do not control the ad content they see, for the simple reason they do not pay publishers (or content owners) to view that content. The Puffin author acknowledges that people also do not sell their attention when they access ad-funded sites, which instead rely on B2B advertising solutions that negotiate between the publisher (seller) and marketer (buyer) of the ad space: "No one will click through to their morning news and act as own auctioneer for any ad they will be seeing."[7]

Because of the different price floor for interest-only ads than contextual-only ads, there will be a higher spread in auction prices than today.

Merely changing the price point for interest ads to win auctions does not reduce the value marketers receive from focusing limited spend on audience-based dimensions (i.e. interest-based advertising). Puffin assumes that the audience information used by marketers to improve their outcomes is generated solely from high-engagement, high-reputation sites. Interest-based advertising leads to "Leakage of ad revenue from high-engagement, high-reputation sites to lower-engagement sites where the same audience appears to be available."[8] However, retargeting is one example where the interest-based information comes exclusively from marketers' own data. Onboarding is another case that illustrates this.

Thus, it is unlikely that prices for context-only ads would increase due to the lack of any increase in value generated from this proposal.

One unwritten assumption of Puffin may be that by reducing the availability of interest-based advertising inventory, marketers would shift their budgets to context-only advertising, thus increasing demand and through this iterative game-theory mechanism increase prices for context-only advertising. Such an assumption relies on another unwritten assumption that marketers would no longer have access to Walled Garden advertising solutions that continue to offer interest-based advertising with contextual targeting and real-time optimization, to message the right audience in the right context at the right time at the right price. These assumptions would rely on Privacy Sandbox eliminating its current first-party and corporate-ownership exemptions (e.g., First Party Sets) to data collection and processing.

As with other Privacy Sandbox proposals that migrate B2B processing to the client, the incremental storage and processing cost for maintaining per publisher, per ad unit, per placement per ad category auctions must be borne by consumers. This may degrade user experience or drain their mobile batteries.

Open Questions

  • What is the impact on web experiences given many client devices do not have the same processing power of the servers that currently select which content to render?
  • How does the local client auction support publisher custom logic, such as factors other than price to select which ad to render?

See Also

References