Are FLOCs really worth it? Criteo investigates

Partager sur facebook
Partager sur twitter
Partager sur linkedin
Partager sur email

While FLoCs are due to be introduced in less than a year’s time, their effectiveness has yet to be proven. And this is what Criteo, a French adtech company, has tried to do, but with mixed results to say the least.

A little reminder on Federated Learning of Cohorts

The idea behind FLoC is to place users in cohorts (audience segments) in order not to have to track their browsing individually as is done with third-party cookies. According to Google, this system respects privacy and above all allows the company to continue to offer an advertising targeting solution. 

A test of limited scope

From 1 June to 13 July, Criteo exploited data from Google’s original test of FLoC, which is claimed to be based on 0.5% of Chrome users (in certain regions of the world only, excluding Europe).

Firstly, Criteo highlights the fact that a majority of the cohorts that were generated (88%) were via a browser from Russia, which is therefore not part of the list of countries for the test. In the remaining 12% that were workable, the company realised that an overwhelming majority were developers using a beta version of Chrome. In the end, therefore, only 0.02% of Chrome users were included in the test, compared to the 0.5% announced.

In addition, Criteo points out the instability of the cohorts generated by FLoC. Indeed, these would only be generated every 7 days, which leads to a lack of concordance and similarity between and within supposedly close cohorts. “This means that overall, we cannot confirm that FLoC is able to capture user behaviour”[1], explains Todd Parsons, chief product officer at Criteo.

It may well be that the promise of efficiency is nothing but a promise.

While Google had announced that its solution could ensure “at least 95% of the conversions” that would have been achieved with traditional targeting via third-party cookies, Criteo, after comparing cookie-based and FLoC-based interest-based targeting, concludes that the recall rate is much lower for FLoC-based targeting.

It should be noted that although specialised domain names will contribute much more to the development of the cohorts (because of the specificity of their audience), they will benefit less than much broader domains (social platforms), which contribute little to the construction of the cohorts.

Considering the results of the Criteo study, the recent postponement of FLoC testing to Q1 2022 is good news, says Todd Parsons: “This means that Google is being more methodical about the next round of testing, and gathering feedback from the industry. We don’t want to be constrained by a feature until it’s validated and proven to work for the consumer, publisher and advertiser”[2].


[1] Roy, Paul. “Privacy Sandbox : Que Faut-Il Retenir Du Test De Floc Réalisé Par Criteo ?” Mind Media, 25 Oct. 2021,

[2] Voir note 1


Sources :

Évaluez votre niveau
de conformité

En quelques clics,
lancez sans engagement
et en toute conformité un
audit flash !

Pour recevoir votre audit flash gratuit et sans engagement, merci de bien vouloir remplir ce formulaire :