Ideas + Opinions

FLoCs vs Cookies – and so the test phase begins

30 April 2021

FLoCs vs Cookies – and so the test phase begins

FLoCs vs Cookies – and so the test phase begins

By Maal Serrate, CTO, Hybrid Theory

In March of 2021, Hybrid Theory issued its technology and solutions roadmap as we move to a world where 3rd Party Cookies are limited or eradicated.  In a widely circulated industry whitepaperHybrid Theory stated its commitment to evolving its proprietary technology to continue to support and advance the efforts of brands and agencies around the world to get the best, most relevant, most timely content to interested audiences. The whitepaper touched on the areas of prospecting, retargeting, relationship/CRM marketing and conversion measurement, as well as the ongoing importance of contextual targeting and first party data.   

In tandem, we are also taking part in industry-wide working groups as we investigate global initiatives for seeing away from dependency of third-party cookies.  One of those initiatives is the Federated Learning of Cohorts – widely known as FLoCThe technology, which uses an algorithm to look at browser history, places users in a group of people with similar browsing histories, or cohorts. 

The Hybrid Theory team has been carrying out tests on this technology and we are now moving, ahead of the industry, to the next stage of our roadmap – the implementation of FLoC IDs on browsers to fully test this prototype in a live environment and test it against our proprietary data set. 

 

What is Federated Learning of Cohorts (FLoC)? 

Federated Learning of Cohorts (FLoC, in short) is a proposal developed under the umbrella of the W3C Improving Web Advertisement Business Group aiming to provide an alternative to 3rd Party Cookies without the need for generating fine-grained browsing profiles of users. 

In both acquisition and conquest, most companies use 3rd Party Cookies to understand user behaviours between different sites visited, correlating the cookie ID between them.  FLoCs place users in a group of people with similar browsing histories, or cohorts. Based on machine learning the technology allows the browser, any browser, to anonymously ‘study’ how users are browsing and then groups them in ‘cohorts’ based on their search behaviour.  

 

FLoC aims to represent privacy-protected interest-based advertising by moving from personalised to aggregated targeting.  

Source: https://web.dev/floc/ 

 

Hybrid Theory FLoC Analysis 

Hybrid Theory will look to validate FloC as a potential alternative to behavioural targeting without the use of 3rd Party Cookies and we will use those learnings to adapt/inform our technology offering to work with cohort interest-based targeting.  

We will be gathering FLoC IDs from our network to understand how Google Chrome is creating cohorts to be able to understand behaviours, motivations and create user categories. And this is where our data scientists come in! They will look to leverage our proprietary data of almost 400M daily events to create custom audiences and lookalike audiences using FLoC IDs, instead of cookie IDs, to validate this test.  

Why test FLoCs? 

It will allow us to validate whether FLoCs are a viable alternative/replacement to behaviour targeting when compared to cookies. Testing it against our proprietary data set will allow us to do a real comparison in a real environment, ensure they are sufficiently private to be made publicly available to the web and help inform how we adapt our product set and offering to include these … or not! 

What this means for agencies and advertisers?   

We are all working toward a common goal: to reach a point where advertisers can find the right audiences for their ads, reach their campaign goals without infringing on privacy. In order to do this, methods for identifying, targeting and measuring are being reviewed. FLoC is one of the initiatives we are exploring to help get us all there. 

We’ll be looking to answer questions that are important to marketeers, such as: 

  • If FLoC targets groups of people, i.e., cohorts, how will we measure product specific interest? 
  • If cohorts are looking to ‘replace’ identity, how will we measure campaign effectiveness? And what will matter most here – viewability or conversion? 
  • If FLoCs only work on the Chrome browser … how private will they really be? 

How many people make up a single FLoC? 

Only FLoCs composed of a minimum of browsers will be shared. Right now, the minimum number of users in a cohort is 2.000. 

How many different FLoCs are we talking about? 

The initial phase shows that 33.872 different cohorts exist. We do know that over time, this number is likely to change.  

Can a user be part of more than one FLoC at the same time? 

No, a user can only be added to a single FLoC. Over time, if the user changes their browsing behaviours, they will likely be placed within another FLoC. 

Can a user opt-out from FLoC? 

Yes. All these initiatives are about ensuring consumers ultimately have control over their data. So users will be able to control this through their browsing settings. In addition, when in private browsing mode, users’ FLoC ID will be filtered out.  

Can a site opt-out from FLoC? 

Yes, a site can also choose not to be included in the user’s list of sites for cohort calculation. 

What about the so-called ‘sensitive categories’ and what are they exactly? 

Testing currently in process is looking to ensure sensitive categories – such as medical websites, and sites with political, adult or religious content – are excluded from these cohorts. According to this paper, a cohort will be filtered out and not shared if the browsing behaviour of its qualifying users has a higher rate of visits to sites with sensitive topics.  

How long will this test phase take and when will we have some findings? 

First of all, we will bring our clients along on this journey! Ultimately, we want to be sure that the FLoC initiative supports your advertising strategies and investments. Outlined below are the phases and proposed timelines for the Hybrid Theory FLoC testing. We will look to provide updates as we meet these milestones: 

  • Q2 2021: gathering and implementation of FLoCs 
  • Q3 2021: analysis and evaluation 
  • Q4 2021: measurement and results 

And finally, what is being done about FLoC testing in territories covered by GDPR law? 

GDPR compliance is an important part of the testing of FLoCs and this work will be announced soon. Google Product Manager Marshall Vale explained that they “are working to begin testing in Europe as soon as possible. We’re 100% committed to the Privacy Sandbox in Europe”

 

References: 

W3C Web Advertising Business Group: 

https://github.com/w3c/web-advertising 

Federated Learning of Cohorts (FLoC) proposal: 

https://github.com/jkarlin/floc 

Evaluation of Cohorts Algorithms for the FLoC API: 

https://github.com/google/ads-privacy/blob/master/proposals/FLoC/FLOC-Whitepaper-Google.pdf 

Measuring Sensitivity of Cohorts: 

https://docs.google.com/viewer?a=v&pid=sites&srcid=Y2hyb21pdW0ub3JnfGRldnxneDo1Mzg4MjYzOWI2MzU2NDgw 

FLoC Article by Google:  

https://web.dev/floc/ 

Hybrid Theory Whitepaper  

https://hybridtheory.com/cookieless-position-paper-march-2021/