2 February 2021
Google’s Privacy Sandbox is a set of proposals to restructure the ad serving ecosystem in order to improve user privacy and provide a sustainable alternative to 3rd party cookies. Below we’ll explain what the main proposals are and how they’ll affect header bidding along with the ad tech industry.
What is Federated Learning of Cohorts (FLoC)?
FLoC or Federated Learning of Cohorts is a proposal from Google’s Privacy Sandbox for serving targeted ads using large groups of users with similar interests. Essentially, by targeting large groups of users instead of individuals and processing user data on-device, FLoC would be able to improve user privacy while enabling targeted advertising.
It is not clear yet that FLoC can be a clean replacement for 3rd party cookies since FLoC currently requires 3rd party cookies to work. The FLoC GitHub (February 2, 2021) documentation states that the following conditions must be met for the FLoC proof-of-concept to log and sync cohort data:
- The user is logged into a Google account and opted to sync history data with Chrome
- The user does not block third-party cookies
- The user’s Google Activity Controls have the following enabled: 1) “Web & App Activity” 2) “Include Chrome history and activity from sites, apps, and devices that use Google services”
- The user’s Google Ad Settings have the following enabled: 1) “Ad Personalization” 2) “Also use your activity & information from Google services to personalize ads on websites and apps that partner with Google to show ads.”
How effective is FLoC?
According to Google’s simulated research, advertisers can expect to see “at least 95% of the conversions per dollar spent when compared to cookie-based advertising”. While the market is still waiting for real-world data to confirm FLoC’s effectiveness, early signs are promising. On 25 January 2021, Google announced that FLoC-based cohorts will be available for public testing through origin trials in March 2021 and advertisers will get to test FLoC-based cohorts in Q2.
How Does FLoC Work?
The FLoC API enables targeted advertising based on a user’s general interests. A machine learning algorithm that operates within the browser determines which interest groups the user should belong to based on the websites the user visits. These interest groups will be made available to advertisers who can select relevant cohorts for their products.
It’s not yet clear whether these interests will have hierarchies or provide targeting on the level of “e-sports fan” or “e-sport fan looking to buy a new monitor”.
FLoC will not replace all functions carried out by third-party cookies. Conversion tracking, retargeting, and fraud prevention are handled by other tools in Google’s Privacy Sandbox.
What is Turtledove?
Turtledove stands for Two Uncorrelated Requests Then Locally Executed Decision On Victory and is an initiative from Google’s Privacy Sandbox. Essentially, this means the browser gets to decide which ad wins, instead of the ad server, and users can’t be identified through ad requests. Turtledove will only work in browsers that agree to support it. It is unclear whether non-Google browsers like Safari and Firefox will adopt it.
Let’s unpack the name in more detail.
Two Uncorrelated Requests - Turtledove
In Turtledove, “two uncorrelated requests” are used to call ads; a contextual ad request and an interest group ad request. The contextual ad request uses the page URL, ad location, 1st party data, and ad size to target users. While the interest group ad request uses a selection of user interest groups, based on the user’s browsing behaviour. To prevent users from being identified, the interest group request doesn’t include the data traditionally passed on to target users such as the URL, user agent, timestamp, or cookies. These requests are made at different times as explained below in order to prevent them from being tied together.
Turtledove Interest Groups
Turtledove enables advertisers to place code snippets on their websites that assign users to interest groups. This interest group information is stored in the browser instead of a 3rd party cookie. As a result, it gives users more control over their data while providing advertisers with a way to target specific user interests with relevant ads. The interest group data that is revealed by the browser with each ad request will be limited to prevent 3rd parties from profiling users.
Locally Executed Decision On Victory - Turtledove
With Turtledove, the browser makes sporadic batch ad requests based on the user’s interest groups, and stores the ads in the browser for later use. When the user navigates to a new page the publisher can choose to just request contextual ads or the publisher can initiate an auction between the contextual ads and the interest group ads stored in the browser. The auction is conducted client-side to separate the contextual and interest group ad requests, making it impossible to match the user to an ID solution.
Overall, this means the browser gets to pick the winning ads along with when and where the ads will be served. In contrast, this decision of which ad gets served is currently made by the ad server based on rules created by the publisher. Critics have called Turtledove a power play from Google since Google is not a neutral third party and would have the ability to tilt ad auctions in the Chrome browser to favor its own ad demand.
What is the difference between FLoC and Turtledove?
Turtledove enables granular retargeting ad campaigns that work on a per user level. In contrast, FLoC provides a way to target large groups of users through common interests.
Will Turtledove from Google's Privacy Sandbox replace the old system?
Turtledove would provide a parallel system for serving ads and will not completely replace the current system for serving ads.
What is SPARROW?
SPARROW or Secure Private Advertising Remotely Run On Webserver, is a development from the TURTLEDOVE project which aims to improve user privacy controls while enabling targeted advertising. SPARROW provides 2 key amendments to the TURTLEDOVE approach:
- Lookalike campaigns: In addition to using user interest groups, SPARROW would allow advertisers to create lookalike campaigns based on interest groups while remaining privacy compliant.
- SPARROW recommends having an independent third party execute the real-time bidding auction instead of the browser. This would protect the user’s data by ensuring advertisers, publishers, and advertising technology partners can’t access personal information while allowing the browser to remain neutral.
Google first hinted in August 2020 that it would be using parts of the SPARROW proposal moving forward with its Privacy Sandbox.
What is FLEDGE from Google's Privacy Sandbox?
The FLEDGE or First “Locally-Executed Decision over Groups” Experiment is a prototype for ad serving from the TURTLEDOVE project. With FLEDGE, interest groups stored by the browser can help advertisers reach specific user cohorts. During 2021, Chrome plans to run an Origin Trial for FLEDGE which you can participate in here.
What is Dovekey from Google's Privacy Sandbox?
Dovekey is a proposal to simplify the bidding aspects of TURTLEDOVE by using the neutral third party auction provider idea from SPARROW (above). According to Dovekey’s authors on GitHub, it has “the benefit of SPARROW bidding even if the gatekeeper server just acts as a simple lookup table.”