Last week Google caused a media storm by announcing that third-party cookies, which are due to be banned from Chrome in 2022, will be replaced with an AI system called Federated Learning of Cohorts (FLoC). FLoC is part of the Privacy Sandbox initiative, a family of privacy-enhancing technologies developed by Google aiming to move the ad-serving process to the browser.
FLoC is a super-tracker that monitors user activity across all sites, stores the information in the browser and then uses machine learning (ML) on the browser to place users into cohorts that share similar interests. This way, advertisers can target groups of people with similar interests without accessing any personal data. Moreover, according to Google’s own tests, FLoC achieves at least 95% of the conversions per dollar spent compared to cookie-based advertising. Publishers, advertisers, rejoice – salvation is near!
Beyond the headline, however, things get messier. At ID Ward, we assess each new cookieless solution from three angles: its ability to support effective marketing in the open web (the commercial angle), its compliance with privacy laws (the legal angle) and its effects on market openness and competitiveness (the competition angle). We ran the test on FLoC and here is what we found.
The inherent limits of the technology
For all of Google’s technological might, FLoC’s performance falls very short of what marketers currently use and expect. Even though the system is still in development, some of its limits are already clear.
FLoC is not a replacement for third-party cookies. FLoC only works in the Chrome browser, leaving cross-browser, cross-device and offline data out of the picture. It has a very narrow focus – to allow interest-based ad targeting – and it does not allow for standard marketing techniques like ad sequencing or frequency capping, to name a few. It is better than nothing, but by itself it is of limited use for marketers.
FLoC’s targeting is *very* blunt. FLoC’s cohorts are relatively fixed and created by Google. This would make it impossible to measure interest in a specific product, it will kill retargeting and the creatives. Something like “show this ad to someone who looked up Adidas sneakers on Amazon at least once in the least 7 days” would turn into “show this ad to sneaker lovers”.
FLoC makes attribution more difficult. If cohorts replace identity, attribution becomes very difficult. How would advertisers measure campaign effectiveness? Viewability metrics would likely take precedence over conversion metrics, and marketing effectiveness would decrease. Other parts of the Privacy Sandbox could fix this by reintroducing identity through the backdoor (for instance TURTLEDOVE could use device data for attribution measurement), but this risks invalidating FLoC’s privacy benefits.
The side-effects for privacy and competition
The second type of concern is that FLoC reinforces Google’s dominant position in the ad market. The problem is not lost on the regulators, and the UK Competition and Market Authority (CMA) and had already opened an investigation into the Privacy Sandbox.
FLoC protects privacy… except from Google. Advertiser’s ability to identify customers across digital properties is at odds with privacy regulations – and replacing IDs with anonymised cohorts of users goes a long way to addressing the problem. But while the industry would only have access to anonymised cohort data, Google would still be able to access the history of the cohorts a user is a part of over time and the raw user data stored in the browser’s cache.
FLoC is a black box. A further problem with FLoC becoming a standard to replace other means of targeting users is that Google will have the ability to tweak the FLoC algorithms to their benefit. The same goes with TURTLEDOVE and other Sandbox proposals. This will be reinforcing their position as player and referee, which doesn’t sit well with a big section of the industry (and arguably with the regulators).
FLoC is unlikely to improve. Those who hope that FLoC will improve over time might have to think twice. First, it is not in Google’s interest to make non-Google ad spend too competitive. Second, broadcasting too much data via the browser would likely lead to de-anonymisation and create massive security and reputational risks for the company. Future improvements will likely be limited to the federated learning algorithm – Google’s black-box intellectual property.
Setting a new path?
Should publishers and advertisers get excited about FLoC? Probably not, or at least not yet. Google’s 95% claim is widely viewed with scepticism, and anything that reinforces Google’s dominant position is unlikely to benefit advertisers in the long term. Similarly, FLoC does not remove the need for publishers to develop strategies to understand and monetise their audiences in a cookieless world, leveraging their first-party relationships. Both publishers and advertisers will have to layer other identity technologies on top of it to power accurate targeting.
There is one silver lining, however. After years looking at cookieless solutions that are neither respectful of privacy nor effective for marketing, Google is betting that the future of digital media is on-device data (personal data stays with the user). This is a big change but one that has been a long way in the making. Two years ago, at ID Ward, we too took the view that the future of digital marketing is on-device data and we built a solution that respects all the principles of FLoC without some of its limitations. More companies will follow down this path over the coming years, realigning the interests of consumers, publishers and advertisers.
Safari and Firefox provide the perfect testing ground before Chrome relegates cookies to the history books. Now is the time for publishers and brands to start testing and scaling cookieless solutions that are compatible with the Privacy Sandbox – not because the industry has to bend to Google’s will, but because on-device data is the best way to reconcile marketing with privacy. Embracing the change is also the only way to protect the market from Google’s thinly-veiled anti-competitive moves.
Dr Mattia Fosci
CEO, ID Ward
ID Ward is an On-device Data Platform (ODP) for cookieless audience targeting. It allows publishers and advertisers unparalleled privacy protection and audience monetisation on Safari, Firefox and other cookieless browsers through its self-sovereign identity and on-device machine learning