Cohort-Based Attribution: What Google's Privacy Sandbox Means for Marketers
Google’s Privacy Sandbox kills cookies. Cohort-based attribution claims to fix it: but fails.
Three platforms, three answers, one budget to defend. Your GA4 export already holds the answer, and the first finding is free and never leaves your browser. Get your first finding free
By Joris van Huët, Founder & CEOUpdated 5 min read
Cohort-Based Attribution Is Not the Answer to Google’s Privacy Sandbox
Google’s Privacy Sandbox just killed third-party cookies. Marketers are scrambling for alternatives, and cohort-based attribution is the latest shiny object. Here’s the truth: it’s a band-aid on a bullet wound. Cohort-based attribution groups users into anonymous buckets, age, location, interests, and measures performance at the group level. Sounds privacy-friendly, right? Wrong. It’s still correlation dressed as causation, and it’s about to leave you with the same garbage data you had before, just aggregated into meaningless blobs.
Cohort-based attribution doesn’t fix that. It just hides the problem behind larger sample sizes. You’re still guessing which ads drove sales, not proving it. And with Google’s Topics API limiting cohorts to broad, interest-based segments, you’re trading precision for privacy theater. The result?
Why Cohort-Based Attribution Fails: The Math Doesn’t Lie
Cohort-based attribution relies on two flawed assumptions:
-
Homogeneity: It assumes everyone in a cohort behaves the same way. They don’t. A 25-year-old in Berlin and a 25-year-old in Munich might share an age bracket, but their purchasing behavior can differ once you account for income, device type, and ad exposure timing.
-
Aggregation Fallacy: It averages out individual behavior, smoothing over the causal chains that actually drive conversions. You’ll never know. You’re left with a blunt instrument that can’t distinguish between correlation and causation.
The Privacy Sandbox’s Topics API exacerbates this. It assigns users to just 350 interest-based cohorts, each containing millions of users. That’s not a cohort. That’s a continent. And when you’re measuring performance at that scale, you’re not measuring anything at all. You’re just vibing.
Here’s the good news: you don’t need cookies to measure incrementality. You don’t even need cohorts.
Causal inference replaces guesswork with proof. Instead of asking, "Did this cohort see an ad and then buy?" it asks, "Did this ad cause this user to buy, compared to what would have happened if they hadn’t seen it?" This isn’t theoretical.
How? By using:
- Holdout Groups: Randomly exclude a subset of users from seeing an ad, then compare their behavior to the exposed group. This isolates the ad’s true impact. No cookies required. Just math.
What Google’s Privacy Sandbox Actually Changes (Spoiler: Not Much)
Google’s Privacy Sandbox is a Trojan horse. It claims to protect privacy while preserving measurement, but what it really does is consolidate power. By limiting third-party tracking, Google forces advertisers into its walled garden, where it controls the data, the cohorts, and the rules of the game.
Here’s what you need to know:
-
Topics API: Assigns users to 350 broad interest cohorts. These cohorts are so large that they’re useless for anything but the most basic brand awareness campaigns. If you’re selling luxury watches, good luck distinguishing between "people who like watches" and "people who buy watches."
-
Protected Audience API: Lets advertisers target custom cohorts, but with severe limitations. Cohorts must contain at least 1,000 users, and you can’t combine them with other data sources. That’s not targeting. That’s throwing spaghetti at the wall.
-
Attribution Reporting API: Provides aggregated, delayed reports on conversions. You’ll get data 2-30 days after the fact, with no user-level granularity. That’s not measurement. That’s archaeology.
The Privacy Sandbox isn’t a privacy solution. It’s a power play. And if you’re relying on cohort-based attribution to navigate it, you’re playing by Google’s rules.
How to Measure Incrementality in a Post-Cookie World
Stop trying to replace cookies. Start measuring what actually matters: incremental sales. Here’s how:
1. Ditch Cohorts. Use Causal Inference.
Cohorts are a relic of the cookie era. They’re a way to group users when you can’t track them individually. But causal inference doesn’t need individual tracking. It needs control groups, natural experiments, and rigorous statistical modeling. That’s not a tweak. That’s a revolution.
2. Run Holdout Tests. Always.
If you’re not running holdout tests, you’re not measuring incrementality. Full stop. Holdout tests randomly exclude a subset of users from seeing an ad, then compare their behavior to the exposed group. This is the gold standard for measuring causal impact. And it works without cookies, without cohorts, and without Google’s permission.
Customer journeys are a fairy tale. They assume that every touchpoint in a user’s path contributed to their conversion. They don’t. Causal inference ignores them.
4. Use Natural Experiments
Natural experiments are real-world events that randomly expose or exclude users from seeing ads. Think ad platform outages, regional blackouts, or even weather disruptions. These events create perfect control groups, and they’re happening all the time. Cohort-based attribution can’t touch that.
The Future of Attribution Isn’t Cohorts. It’s Proof.
Google’s Privacy Sandbox is here. Third-party cookies are dead. Cohort-based attribution is the industry’s desperate attempt to cling to the past. But the future of attribution isn’t about grouping users into anonymous buckets. It’s about proving what actually drives sales.
Causal inference is that proof. It doesn’t need cookies. It doesn’t need cohorts. It doesn’t need Google’s permission. It just needs math, rigor, and a willingness to challenge the status quo.
They’re not waiting for Google to fix attribution. They’re fixing it themselves.
Ready to join them? See how causal inference works for ecommerce brands.
FAQs
What is cohort-based attribution, and why is it flawed?
Cohort-based attribution groups users into anonymous segments and measures performance at the group level. It’s flawed because it assumes homogeneity within cohorts and relies on correlation, not causation.
What are the limitations of Google’s Privacy Sandbox for attribution?
Google’s Privacy Sandbox limits cohorts to broad segments (350 interest-based groups), delays attribution data by 2-30 days, and restricts custom cohort sizes to 1,000+ users.
Sources and Further Reading
Related Articles
Frequently asked questions
What is cohort-based attribution, and why is it flawed?
Cohort-based attribution groups users into anonymous segments and measures performance at the group level. It’s flawed because it assumes homogeneity within cohorts and relies on correlation, not causation.What are the limitations of Google’s Privacy Sandbox for attribution?
Google’s Privacy Sandbox limits cohorts to broad segments (350 interest-based groups), delays attribution data by 2-30 days, and restricts custom cohort sizes to 1,000+ users. This reduces measurable conversions and eliminates user-level granularity, making precise incrementality measurement impossible without causal inference.
Go deeper: Incrementality testing, explained.
Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.
Have an audience of ecommerce operators? Earn 20% recurring commission by recommending the €99 read.
Keep reading
Terms in this article
- Attribution ReportAttribution Report shows which touchpoints or channels receive credit for a conversion. It identifies which campaigns drive desired actions.
- Causal ChainA Causal Chain is a sequence of events where each event causes the next, leading from an initial cause to a final effect.
- Causal InferenceCausal inference is the set of methods for estimating how much an outcome changes because of a cause, rather than alongside it. It works by comparing what happened with an estimate of what would have happened without the cause.
- Control GroupA control group is the part of an experiment that does not get the treatment being tested, so its outcomes stand in for what the treated group would have had without it. In marketing it is the audience, or the set of regions, held back from a campaign.
- Customer journeyCustomer journey is the path and sequence of interactions customers have with a website. Customers use multiple devices and channels, making a consistent experience important.
- IncrementalityIncrementality is the part of an outcome, such as orders or revenue, that a marketing activity caused: the extra results compared with what would have happened without it. It is a causal effect, not the number of conversions credited to a channel.
- Natural ExperimentNatural Experiment is an empirical study where experimental and control conditions are determined by nature or external factors. This estimates causal effects when randomization is not feasible.
- Third-Party CookieThird-Party Cookie is a cookie set by a domain other than the one a user currently visits. These cookies track users across sites for advertising.