There was an unexpected error authorizing you. Please try again.

Adopting SDA in a Privacy-Centric World


First released over a year ago, seller-defined audiences created a mechanism for sellers to provide standardized information about their audience to ad buyers.  In a world with declining user-level signal availability, this was proposed as a solution to retain addressability for the buy-side of ad tech.  This specification, along with standard audience and content taxonomies, allowed an elegant solution for buyers to transact on a publisher’s  1st-party data in a scalable way.


At the time, a challenge was recognized, which was that if these new SDA signals were sent in bid requests along with user-level signals (IP, user id, etc) then consumers of such requests could store and build profiles on these users.  The specification acknowledged this with the statement “Accountability of supply chain participants to minimize commingling of cohort signals with other identifiers…” (page 20).  However, this specification assumed a world in which user-level signals no longer existed, which has not come to pass given both Chrome’s delayed changes and the rise of various alternative IDs and other user-level signals.  


While a clear solution to the above directive would be to strip all user-level information from bid requests, doing so would not be advantageous to sellers given how critical those signals are to monetization.  They’d be better-off not adapting SDA and continuing to send user signals, which indeed is what has largely transpired.

In web environments without cookies, simply removing any user-level signals and sending SDA may be revenue-optimal.  But for many requests cookies, UIDs, and other user-level signals are still available (even on browsers without 3rd-party cookies).  How can sellers send SDA AND user-level signals?


Our new SDA implementation guide outlines a path to doing both without sacrificing privacy by obfuscating the SDA signals via deal IDs.  Already, sellers use deal IDs to signal many things to buyers, including audience membership.  Sellers can package SDA signals into deal IDs to signal to buyers.  As long as these deals are aggregated from multiple SDA segments of interest to a buyer, are buyer-specific, are not included in deal catalogs, and are not named descriptively (attributes that are already true of many deals today), the buy-side cannot ascertain the audience data used to create them.  If deals are obfuscated in this way, then the SDA signals themselves stop at the seller (typically an exchange), and the deal IDs and available user-level signals can flow to the buy-side.  This diagram illustrates the flow:


To be concrete, here is an example data flow:

  1. Publishers gather audience data through logins, surveys, inferred log data, etc, and store it in 1st-party cookies or other mechanisms
  2. Publishers make this data available via SDA signals to trusted exchange partners
  3. An exchange is working with a buyer on a campaign and works with the buyer to understand the audience they are seeking, for example those interested in home improvement
  4. The exchange finds all the SDA segments related to this interest, and when a request comes in with one of those segments (from a set of multiple publishers), maps it to a deal ID
  5. That deal ID is communicated to the buyer and they target the deal
  6. As requests come in with those SDA segments, the exchange adds the deal id to the request, but does not pass on the SDA signals themselves
  7. User-level signals, to the extent they exist, can continue to be passed


No solution is perfect; let’s discuss some drawbacks and considerations.

  • This solution requires a trusted intermediary to aggregate, package, and sell the deals on behalf of publishers as this solution will work best when intermediaries package segments from multiple sellers.  Publishers have to trust their exchanges to do this correctly.
    • Scaled publishers could of course do this themselves through directly-sold PMPs or other mechanisms
  • This does sacrifice the elegance of a standardized taxonomy flowing throughout the ecosystem that everyone can understand since we are obfuscating the SDA signal at the intermediary layer; however, retaining the taxonomy flow would require removing user-level signals which would be a tremendous harm to monetization


The key next step is that sellers need to begin gathering and storing 1st-party data.  It’s important to note this does not require logged-in users.  Publishers can use 1st-party cookies and other mechanisms to create profiles about their audience.  Publishers will then need to work with their trusted exchange partners to pass this data through and monitor their partners success.

Exchanges need to implement mechanisms to ingest, package, market, and sell media based on this data.

The industry is still in transition, with a mix of cookie-based and non-cookie-based user-level signals as well as a growing segment of inventory without user-level signals.  We need pragmatic solutions like Seller Defined Audiences to bridge us to a more stable privacy-safe future that maintains a viable business model for journalists and content creators.


Curt Larson