How well does Google’s replacement tech for third-party cookies work? So, so (so far).
On Tuesday, Google Ads , Google’s third-party ad tech, published the results of an experiment designed to isolate the contribution of interest-based audience solutions, including the Topics API, first-party publisher IDs and contextual data.
The experiment included a test of the Chrome Privacy Sandbox proposals and of Google’s own privacy-focused advertising tools.
Google has been working on what it called interest-based audience products as a long-term way to mitigate the loss of online tracking data, according to Dan Taylor, Google’s VP of Global Ads, who briefed a group of reporters about the ad experiment on Monday.
The Google Ads test was conducted from the perspective of third-party ad tech providers (which Google refers to as “ATPs,” because acronyms are a way of life).
This study follows other recent experiments in the Chrome Privacy Sandbox led by ATPs (sorry), including Criteo and RTB House, both of which have run into snags.
And now Google Ads shares their frustrations with testing Privacy Sandbox proposals.
And Google Ads is in a similar situation as third-party ad tech, which is increasingly beholden to how browser makers define privacy. After all, a growing number of users choose their browser with privacy in the forefront of their minds.
“It turns out 80% of people are now concerned about the state of their online privacy, and almost half are turning away from services due to privacy concerns,” Taylor said. “So that’s scary.”
The good
The Google Ads experiment was encouraging for data-driven advertising without third-party cookies, Taylor said.
The test involved a control group of Chrome audiences that were targeted using third-party cookies and a test group targeted using only interest-based signals: site-based contextual data, publisher first-party IDs when available and data from the Topics API. (Topics classifies users by interests like “Fashion” or “News” based on their browsing history.)
“The campaigns using privacy-preserving signals to reach users maintain a pretty high range of fidelity relative to third-party cookie-based performance,” Taylor said.
Clickthrough rates declined by less than 10% without third-party cookies, compared to third-party cookie targeting, and the conversion rate per dollar spent was down between 1% and 3% for the test group.
That’s not exactly a call-to-arms, but it does represent progress for Google’s privacy product toolkit. Google wants the Topics API, which in the past has been tested as a standalone product, to be incorporated instead as part of a patchwork of data which, in this case, also includes publisher IDs and contextual data.
The bad
Still, there are a lot of caveats to call out here.
Although the Google Ads study does add to the body of Privacy Sandbox experimentation, it also shows the long road ahead for the Chrome group before its ready to meet its own Q3 2024 deadline for third-party cookie deprecation.
And this test was limited in many ways – so many ways, in fact, that it’s difficult to pull much from these results other than lukewarm optimism.
The A/B trial campaigns ran for just five weeks, which isn’t a sufficient window for some product sales, Taylor said. Five weeks also isn’t enough time for Google’s machine learning optimization to hit its stride.
The tests also only ran in Chrome across display campaigns. And, although Google Ads avoided third-party cookies for targeting purposes with the test group, both the test and control campaigns incorporated third-party cookies for attribution, retargeting and frequency capping.
Still, according to Taylor, the test gave Google “a good understanding” of how audience targeting for interest-based segments, such as “in-market” and “affinities” would perform in practice compared with cookie-based audiences, Taylor said.
But if Google’s privacy-preserving ad products are to stand on their own, they eventually must do so without third-party cookies across an entire campaign, including measurement.
The asks
To achieve this, Google Ads has a few Topics API-related requests for the Chrome team.
One suggestion, echoing other third-party vendors, is to assign Topics granularly by URL rather than host name. (This means that ESPN visitors, for example, get assigned a general “sports” Topic, when specific articles might tag some of those same visitors more directly as “New York sports fans”).
Google also suggests removing Topics that are easily divined from a web page itself. For instance, ESPN shouldn’t get “sports” as a category at all. Because, duh.
Instead, the Google Ads team is calling for the Topics API to favor more granular and commercially valuable categories. Niche sites, like a camera enthusiast page, for instance, may not generate enough traffic to merit a Topic assignment, because the API favors more trafficked sites. But that niche site’s “camera enthusiast” Topic assignment holds more value than someone lumped into far broader “news” or “sports” segments.
Another headwind Google Ads acknowledged is that as third-party cookie deprecation unfolds over the next year and as other advertising IDs diminish across the web, there will be less precise classifying data about users.
The prospecting pool of “shoe shoppers” may still flow, but there will be fewer and fewer known “Jordan enthusiasts” in the mix, resulting in less effective conversion rates, per one example from the Google Ads test results.
The “winner”
But challenges aside, the the standout in the Google Ads study was … the Google suite of machine-learning ad products. Quelle surprise.
Google’s machine learning products, such as Performance Max, Automated Bidding and modeled conversions, give Google Ads more control over targeting and measurement, while advertisers have less visibility and reduced access to campaign data.
But, as Taylor noted, Google’s machine learning products do help advertisers achieve effective targeting at scale, because the Google engine is “filling the gaps” lost to third-party cookies.
Machine learning unlocks new audiences because advertisers are bound by the limits of their own data and by privacy-related restrictions. Without a machine learning boost, advertisers end up targeting close lookalikes or retargeting their own data. It’s hard to find new, non-obvious prospects when browser-based privacy controls prevent connections and inferences about users.
Not that machine-learning ad products are without pitfalls. Buyers are concerned about having to hand over the keys to a platform for targeting and self-attribution. But that isn’t slowing Google down.
“This last insight in particular is what I’m passionate about,” Taylor said. “We are all-in on utilizing Google’s AI in our advertising products.”
