In the latest update on its plans to replace third-party cookies for advertising, Google said tests on one particular proposal look promising.
Google planned to share some new findings showing the effectiveness of its “Federated Learning of Cohorts” proposal that’s part of the Chrome browser’s “Privacy Sandbox” in a blog post running Monday. The “Sandbox” is an initiative launched in 2019 to find alternatives to the cookie while mitigating the impact on publishers and other players. In Google’s words, it was about finding a solution that both protects user privacy and lets content remain free available on the open web.
Not long after announcing the initiative, Google said it would be ending support for third-party cookies, which fuel much of the digital advertising ecosystem, in its Chrome browser within two years of January 2020.
Chrome engineers have been working with the broader industry, including with web standards organization W3C, on ideas in the Sandbox that have been proposed by Google and other ad tech players. What’s likely to result is a number of these ideas moving forward, Google says.
“This is one proposal,” Chetna Bindra, group product manager for user trust and privacy at Google, told CNBC of the “FLoC” progress. “It is absolutely not the final or the singular proposal to replace third-party cookies … There won’t be one final API that will go forward, it will be a collection of them that allows for things like interest-based advertising, as well as for measurement use cases, where it’s critical to be able to ensure that advertisers can measure the effectiveness of their ads.”
Bindra said the company is “extremely confident” about the progress on the proposals and the tests thus far.
Google’s post Monday says that test results show that FLoC (pronounced like a flock of birds, in keeping with a number of bird-themed proposals like “Turtledove” and “Swallow”) is “an effective privacy-focused replacement signal for third party cookies.” It says advertisers can expect to see at least 95% of the conversions per dollar spent when compared to cookie-based advertising.
FLoC would essentially put people into groups based on similar browsing behaviors, meaning that only “cohort IDs” and not individual user IDs would be used to target them. Web history and inputs for the algorithm would be kept on the browser, with the browser only exposing a “cohort” that holds thousands of people.
“We are really seeing that one of these first Sandbox technologies for interest-based ads is literally nearly as effective as third-party cookies,” Bindra said. “There’s certainly a lot more testing coming. We are very keen for advertisers and ad tech to engage directly.”
Bindra said these cohorts, which could include people who have behaviors like an interest in gardening or in rock music, would still allow for targeting based on those interests. Instead of targeting on an individual level, though, this would target groups.
“The difference will really just be that now they’re no longer tracking every user across the web. There really is that notion of privacy for those users that are now clustered within a cohort,” Bindra said.
She added that the figures from the tests of FLoC should be reassuring to publishers. Next, Chrome will make the cohorts available for public testing with its next release in March, and expects to begin testing FLoC-based cohorts with advertisers in Google Ads in the second quarter, the blog post says.
Myles Younger, a senior director at the global data practice at MightyHive, said the Sandbox proposals are all getting at “how can we build new features into the Chrome web browser to simultaneously solve for user privacy and the death of the third-party cookie while preserving brands’ ability to advertise effectively.” He spoke before Google’s latest findings were released.
One question is whether players are actually going to use it.
“I’m not sure it’s something Google is able to just flip a switch and turn it on,” he said. “Publishers have to use it. People have to start using this system. [Google] needs to prove it works.”
Paul Bannister, chief strategy officer at CafeMedia, said advertisers and publishers do have some fear of the unknown as it pertains to what comes next.
“I think we all want to believe this will be good and we all want to get to a place where users have more privacy and the web works better,” he said. But given how complicated and technical the process is, it’s unclear what will really happen next.
He said there’s some fear that these kinds of actions could advantage the “walled gardens” of companies like Facebook, and away from advertising on the open web.
U.K. antitrust authorities also have an eye on the plans and are investigating whether the plan to remove third-party cookies from Chrome could hurt online ad competition. The Competition and Markets Authority is looking into whether Google’s plans could cause advertisers to shift spend to Google’s own tools at the expense of its competitors.
In an emailed response, Bindra said, “The Privacy Sandbox has been an open initiative since the beginning and we welcome the CMA’s involvement as we work to develop new proposals to underpin a healthy, ad-supported web without third-party cookies.”
Some privacy advocates are also skeptical of the “FLoC” approach. The Electronic Frontier Foundation wrote in 2019 that these cohorts could be used in harmful ways, letting discriminatory advertisers identify and filter groups representing vulnerable populations.
“A flock name would essentially be a behavioral credit score: a tattoo on your digital forehead that gives a succinct summary of who you are, what you like, where you go, what you buy, and with whom you associate,” EFF staff technologist Bennett Cyphers wrote in the blog post. “The flock names will likely be inscrutable to users, but could reveal incredibly sensitive information to third parties.”
Whether the machine learning would create cohorts based on health issues or low-income status or other sensitive attributes is question for some.
“It can potentially do very creepy and arguably illegal things,” Bannister said. “How will Chrome protect against that?”
Google said in documents that its analysis evaluates whether a cohort may be sensitive without learning why it is sensitive, and said cohorts that reveal “sensitive categories” like race, sexuality or personal hardships were blocked or the clustering algorithms were reconfigured to reduce the correlation.
Google added that it’s against its policies to serve personalized ads on these sensitive categories.
Nominations are open for the 2021 CNBC Disruptor 50, a list of private start-ups using breakthrough technology to become the next generation of great public companies. Submit by Friday, Feb. 12, at 3 pm EST.