How Google Ads delivered comparable results using Privacy Sandbox APIs?

For nearly three decades granular cross-site tracking of user activity—third-party cookies—have been central to the personalization of digital advertising. That era will soon end. As we inch closer to the cookieless cliff, new methods are emerging for how to replace them and achieve comparable results. 

For example, with third-party cookies it’s possible to record that a person visited an airline website where they searched for a trip to Cannes in June. The same person also went to a luxury hotel booking website searching for accommodation. They’ve also browsed a few websites of top advertising companies and agencies. All this information added together suggests that this person is likely an executive attending the Cannes advertising festival. It can help inform algorithms to display ads with high-end brands, or high-ticket experiences such as yacht rentals—goods and services they’re likely interested in.

To achieve similar personalization without third-party cookies, some members of the industry are focusing on inventions trying to replicate third-party cookie mechanics using external identifiers, also known as universal IDs. This segment of cookieless solutions is generally frowned upon by regulators, publishers, user privacy advocates, and browser developers, including Chrome, so let’s see which cookieless methods satisfied targeting use cases in a more privacy-friendly way.

First-party data is key, but how to use it well?

Solutions after phasing out third-party cookies aim to keep user privacy intact while also keeping ads personalized. That’s why the future will largely depend on capturing first-party data and grouping users, so that when they leave one publisher’s properties, they can’t be re-identified on another site. 

There are several methods to capture and pass first-party data, which can be further enriched with certain privacy-preserving signals. Such mapping will become more powerful when a trusted technological partner is there to help. 

But there’s little sense in fixating on one perfect tool that’s going to fix all our future cookieless problems.

Google study suggests a portfolio of tools necessary

Portfolio of tools more effective than singular solutions, Google study finds

To visualize possible results with using privacy-preserving tools in comparison to third-party cookies, Google published the results of its interest-based advertising (IBA) testing using some cookieless-dedicated technologies. The experiment relied on comparing the IBA performance—a mixture of contextual signals, Topics API signals, and first-party IDs—to third-party cookies. Frequency capping and measurement were provided through the use of third-party cookies in both the test and the control group. Many of the outcomes were encouraging as Google only observed a 2-7% decrease in Google Display Ads advertiser spend (IBA with privacy-preserving signals against third-party cookies), along with 1-3% decrease in conversions per dollar, and click-through rates within 90% of the legacy technology. 

The industry received the article very well as it was Google's first meaningful attempt to quantify the effect of using technologies other than third-party cookies. However, even though the test group employed multiple privacy-enhancing solutions, results were still slightly worse than with third-party cookies alone. We can expect that if Google had removed one (or more) tool from the test group mixture, this would have dramatically worsened the results. 

At the same time, it’s difficult to say what the results will be for specific campaigns. For example, in a broad targeting scenario delivered for, say, Coca-Cola, the decrease in performance may be minimal. On the other hand, in a precise targeting scenario for niche brands, like a boutique hotel for Spanish-speaking students in Manchester, it might be much more difficult to achieve satisfactory results with tools used by Google Ads. This proves that the future belongs to mixtures of various cookieless solutions, rather than one-size-fits-all remedies. 

Another test, this time performed only on partitioned first-party identifiers by a publisher—Graham Media, provided with another set of promising results. Ads that used survey-based audiences encoded into PPID brought a 29% larger click-through rate than those with no PPID, on top of which the publisher’s CPMs have increased by around 6% since PPID implementation. Graham Media also emphasized that the company doesn't intend to use any external identifiers that can be traced back to an individual, which would replicate the flaws of third-party cookies.

What happens next?

It’s much easier to hold on to the status quo, but with enough cooperation and sound strategy on which tools to use and how, the vast majority of personalization use cases can be satisfied with cookieless solutions, without the need to sacrifice user privacy. 

Let’s also not forget that a large portion of third-party cookie alternatives provide a very high certainty that advertisers reach a relevant audience (in terms of Protected Audience API its as large as 100%), while match rates of commonly used cookie matching tend to oscillate around 50% or even less, according to Joey Trotz, head of product for the Privacy Sandox Ecosystem.

We’re nearing a very intense and exciting time, as Privacy Sandbox’s General Availability (GA) stage is coming in July 2023. During the GA, MarTech companies will be able to test the Privacy Sandbox at scale and then provide feedback about which functionalities fail to deliver larger traffic. In Q1 2024, Chrome will remove support for third-party cookies from 1% of its results. 

This will provide a reliable cookieless world simulation, in which one could meaningfully test performance of various cookieless solutions. Over the next few months we can expect plenty of reports from countless tests, claiming that one cookieless tool is better than another. As the dust settles, we all need to maintain a healthy dose of cynicism, and be mindful of different test methodologies, constraints, targeted use cases and environmental factors, which might impact those tests one way or another.