FLoC Federated Learning of Cohorts

Federated Learning of Cohorts (FLoC) is one of Google’s new technologies being tested for roll out to enhance privacy on the web. The aim is to give people more transparency and greater control over how their data is used.

Essentially, FLoC is a privacy-preserving technology Google is developing for interest-based ad selection that will replace third-party cookies.

This is done by placing users in large groups, or cohorts. These groups aren’t based on who the individuals are, but rather on their collective interests based on their browsing history. Cohorts are not a collection of people. They are groupings of browsing activity.

In January 2020 Google announced that it would eliminate third-party cookies from Chrome by 2022. (Apple’s Safari and Mozilla’s Firefox are already blocking third-party cookies by default.)

In the beginning of 2021, Google rolled out FLoC to a small number of Chrome users. According to Google’s Privacy Sandbox timeline, the origin trial for the initial version of FLoC ran until the third quarter of 2021.

However, in June 2021 Google published an announcement that it is delaying its plans to phase out third-party cookies in Chrome to 2023. The first transition phase for FLoC is due to start in the fourth quarter of 2022 with transition phase 2 being planned for quarter 3 of 2023.

• Show ads to users who frequently visit advertiser’s sites or who show interest in relevant topics.
• Predict how likely it is for a user to convert. This is done using machine learning – this is useful in ad auction bidding when it comes to the behaviour metric.
• Recommend content to users. For example, when a sports podcast page on a radio station site becomes popular with visitors from certain cohorts, that content can be recommended to other visitors from those cohorts.

1. The FLoC service creates cohort segments.
The FLoC service creates a multi-dimensional mathematical representation of all potential web-browsing histories. It then divides this model up into thousands of segments. Each segment represents thousands of similar browsing histories. Each of these segments is given a cohort number.

2. The user’s browser calculates the user’s cohort.
Using the FLoC model’s algorithm, the browser works out which cohort segment corresponds most to its own browsing history. It then assigns a cohort number to itself.

3. Advertisers observe cohort activity. Sites that pay for advertisements can include code on their own websites to collect and provide cohort data to their adtech platforms. So when a user visits a site that sells products, the site asks the browser for its cohort. The site then records the cohort’s product interests and shares this data with adtech platforms. (Adtech platforms are companies that provide tools and software to deliver advertising.)

4. Publishers observe cohorts of their site visitors. When the user visits a publisher (a site that gets paid to display ads), the site observes the cohort and shares interests shown with adtech platforms.

5. The adtech platform selects an ad to display to the user. Combining the data received from the advertiser and the publisher, the adtech platform calculates which ad would be most relevant to the user’s cohort.

The user’s browser recalculates its cohort regularly on the user’s device. This is done without sharing individual browsing data with any third party.

Need help with getting your products in front of the right audience? Enquire about our Google Ads Management service.

• Businesses rely on targeted advertising to drive traffic to their sites.
• People generally prefer to see relevant and useful ads.
• Relevant ads are more effective in bringing in business to advertisers and more revenue to the publishing websites.
• Advertising space is more valuable when it displays relevant ads.
• Relevant ads help fund content creation that benefits users.

Targeted advertising that currently relies on techniques such as tracking cookies can reveal users’ browsing history across the web to advertisers or ad platforms. People are becoming more concerned about the privacy implications of this.

FLoC enables ad selection without sharing the browsing behaviour of individual users. This new technology enables ad selection that better protects privacy. During its initial test rollout, Google ran FLoC tests on affinity Google Audiences and in-market audiences. Google claims that advertisers can expect to see at least 95% of the conversions per dollar spent when compared to cookie-based advertising. It seems that FLoC can be a viable alternative to using third-party cookies.

The key differences between FLoC and third-party cookies:

FLoC Third-party cookies
> The user’s browsing history is stored locally on the browser. The FLoC service does not store any user data.
> Collects information about interest-based user behaviour only.
> It uses the browser’s cohort number to target advertising to the user.

Third-party cookies
> User information is distributed or uploaded outside of the user’s device.
> Collect personal identifiable information such as age, origin, gender, and user behaviour data.
> Use the individual’s user ID to target advertising.