To understand Googleâs Federated Learning of Cohorts (FLoC), we need to rewind a little bit. In early 2020, Google announced that it would be phasing out support for third-party cookies on its Chrome browser by 2022, effectively following in the footsteps of Appleâs Safari and Mozillaâs Firefox browsers. The impact of this, of course, is significant as it marks dwindling support for one of the internetâs most ubiquitous user tracking techniques. Without third-party cookies on these three browsers, approximately 87 percent of users online can no longer be tracked with his method.
Arguably, Googleâs strategy canât be faulted on the face of it. The move is a step in the right direction in the eyes of privacy advocates (for the most part) and sets a new precedent for what the future of user identifiers is set to look like. To ensure that advertisers arenât left without any tools at their disposal, Google has been working on its Privacy Sandbox since 2019 in order to power a more privacy-oriented internet. This sandbox aims to establish open standards for tracking users while protecting their right to privacy and FLoC is part of one of the initiatives in the Sandbox.
FLoC allows for targeted advertising and audience profile creation to take place without collecting an individualâs browsing history. Instead, FLoC enables advertisers and AdTech firms to target a âcohortâ (or a âflockâ) of similar users. These cohorts are developed, updated, and refined by browsers based on user browsing behavior â those with similar browsing habits are grouped together and these cohorts are updated over time as users navigate the web. By leveraging machine learning algorithms, the browser can develop a flock based on specific input features (e.g. the URL of the visited sites or the content on the page). FLoC is unique as it ensures that all browser history remains in local storage â the browser only ever ârevealsâ the generated flock that the user is part of. However, the success of FLoC depends on the quality of these clusters. Labels need to be suitable for machine learning and clusters need to be significantly distributed and sizable so that they accurately represent the interests of the group as a whole, rather than a person.
While real-world testing of FLoC has already begun, Google recently announced that it could not run in countries where the GDPR and the ePrivacy Directive are in effect due to privacy concerns. Despite this being an alternative that supposedly offers greater privacy for users, some regulators are arguing that it still isnât enough.