Upcoming: Gaming & Advertising Guide, Learn More>

Does Amazon Have the Power to Hamstring FLoC?

Tim Cross  08 July, 2021

Google announced at the beginning of the year that it is backing its Federated Learning of Cohorts (FLoC) API as an alternative to third-party cookies for ad targeting at the beginning of this year. FLoC monitors users’ browsing habits and places them into large anonymised groups of users with similar browsing habits (cohorts) which advertisers can then target.

Over the past few months, a trial version of FLoC has been released for testing, helping the industry to get a feel for how it works, and to feed back on what needs improving.

But we’ve seen numerous reports that some companies aren’t playing ball, blocking the FLoC API from gathering data. Last month Digiday reported that Amazon has put code on a lot of its owned and operated websites, preventing FLoC from harvesting data. This looked like something of a power play from the e-commerce giant, shielding its own valuable data from Google’s prying eyes.

This raised an intriguing question. Will Amazon, or anyone else in the industry, have the power to hamstring FLoC?

The answer is yes and no.

A shift to opt-in

On a technical level, any website owner will be able to prevent FLoC from working on their domains. For the ongoing FLoC trials, Google has taken an opt-out approach, where any website may be included if Google detects that it’s using ads-related resources, or if it is actively choosing to use the FLoC API.

But Google has told website owners that they can opt-out of these trials by setting a specific permissions policy on their page. A Google spokesperson confirmed to VideoWeek that domains can not only prevent FLoC from scraping any data to help build cohorts, but they can also prevent FLoC cohorts from being used to target ads on their page. So if a publisher or ecommerce site doesn’t want to work with FLoC, they have the power to shut it out completely if they choose.

Zach Edwards, founder of analytics and optimisation firm Victory Medium, told VideoWeek that he expects if and when FLoC is fully rolled out, it will switch to an opt-in model instead, where domains actively choose to enable the API.

Edwards pointed to a ruling from the Federal Trade Commission earlier this year which required photo storage site Everalbum to delete algorithms and models which had collected and processed user data to train their facial recognition AI, without telling users they were participating in research. He said this same logic would apply to Google’s use of FLoC, were it to remain opt-out.

So Google is well aware that any website may choose to block FLoC. The next question is how big of an impact this will have.

What data is used to build cohorts?

Obviously if every website chose to block FLoC, it would be dead in the water. The API would have no data to be able to build cohorts, and no ability to use cohorts to target ads.

This scenario seems unlikely. Advertisers don’t want to lose the ability to target ads based on user behaviour, so publishers will likely be pressured to use any tools which allow them to do so.

Companies with stores of their own first-party data, like Amazon, are the only ones for whom blocking FLoC would make much sense. They won’t be reliant on FLoC for user targeting. And blocking FLoC means more of their data is kept siloed within their own walls.

Losing Amazon’s data will be a blow. “Amazon is the largest western company with the largest product catalogue,” said Victory Medium’s Zach Edwards. “So their decision to block FLoC across all of Amazon.com and numerous subsidiaries of theirs will dramatically reduce the quality of web cohorts.”

But the extent of the damage will depend on the sorts of data which FLoC uses to build its cohorts in its final form.

A Google spokesperson told VideoWeek that in the ongoing trials, FLoC only uses domain names to build its cohorts. So while Amazon obviously has tonnes of valuable data around which products users are looking for, most of this data wouldn’t be used to build cohorts anyway, since it’s not included in the domain name.

But the spokesperson said that Google will review the data used in building cohorts once the trial is finished. If more granular data web page data is used to build cohorts, then Amazon’s data would be more of a loss. Even a shift to using URLs rather than domain names would be significant. An Amazon URL could, for example, reveal users who have searched for ‘PlayStation 5’ – a potentially valuable datapoint for building cohorts.

And obviously Amazon’s ability to block ads sold on its pages from using FLoC has a big impact too. The more big companies opt-out of FLoC, the less valuable it becomes as a tool for targeting users, since it applies to a smaller share of an advertiser’s media plan.

But as things stand, Amazon’s move is an inconvenient pain for Google, but not a FLoC killer.

Zach Edwards, however, said that Amazon’s impact could be even bigger if it’s decision to block FLoC causes others to follow suit.

“As more publishers and website owners realize that they are providing training data for a Google ads product that has murky privacy protections, it will be likely that more websites will follow Amazon’s lead, and block Google’s newest unsafe advertising product,” he said.

And he said that if more e-commerce stores choose to block FLoC, the entire API could lose its value. “As soon as we see Alibaba or any Chinese ecommerce retailers following the model of Amazon, that will be a sign that FloC will never reach global scale. It will slowly lose traction with publishers who are paying attention to their own value, and partners attempting to exfiltrate audience value without paying for it.”

Go to Top