Marketing Attribution is Dead. Here's What Replaced It.

Since November 2023, GA4 has offered three attribution models: data-driven, paid and organic last click, and Google paid channels last click; first click, linear, time decay and position-based are no longer available. What replaced rule-based attribution isn't a better rule. It's measuring what a channel caused: hold spend back somewhere and compare, or estimate the counterfactual from data you already have.

For years, attribution tools sold the idea of tracking every touchpoint and giving each one its precise share of credit. What you got was conflicting dashboards, inflated ROAS and no way to tell what actually made a customer buy. Consent banners, browsers that block third-party cookies and ad platforms that keep their data to themselves have since punched holes in every click path, so traditional marketing attribution now hands out precise credit for an incomplete picture.

For a Dutch beauty or fashion brand, that means budget decisions made on numbers that can't all be true at once. The way out is to stop asking which touchpoints were present and start asking what would have happened without each one; the death of attribution piece covers how that shift played out.

Why Your Attribution Model Is a House of Cards

Your attribution model is a house of cards because it's built on correlation, not causation. It records which touchpoints were present, not whether they made a difference. Causal analysis asks the second question; rule-based models answer the first and present it as the second.

For over a decade, digital marketing ran on a simple premise: track everything. That produced a zoo of attribution models, each claiming to pinpoint which channels deserve credit for a conversion. Every one of them, from first-touch to W-shaped, splits credit by a rule someone chose, not by what each channel did.

First-Touch and Last-Touch: The Original Sinners

First-touch attribution gives 100% of the credit to the first channel a customer touched, while last-touch gives it all to the last. Imagine a shopper in Amsterdam discovers your sustainable fashion brand via a blog post, sees a retargeting ad on Instagram a week later, and finally buys by searching your brand on Google. First-touch credits the blog and ignores what social and search did. Last-touch credits Google and ignores where the shopper found you. Both tell a story with a beginning or an end, but no middle.

Linear and Time-Decay: Spreading the Inaccuracy Evenly

To fix this, models like Linear and Time-Decay emerged. A linear model divides credit equally across all touchpoints. In our example, the blog, Instagram ad, and Google search each get 33.3% of the credit. This appears fair, but it wrongly assumes every touchpoint is equally valuable. Was the initial discovery on the blog really as important as the final, high-intent search? A linear model has no answer.

The time-decay model tries to be smarter by giving more credit to touchpoints closer to the conversion. The Google search gets the most credit, the Instagram ad less, and the blog the least. This is an improvement, but it is still an arbitrary assumption. What if the initial blog post was so compelling it was the true cause of the purchase? The model has no way of knowing. It’s a guess dressed up as data science.

U-Shaped and W-Shaped: Elaborate Fictions

Then came more complex models like U-Shaped (crediting the first and last touch) and W-Shaped (crediting the first, middle, and last). They're more elaborate ways of guessing, with an air of sophistication and the same blind spot: they track events, not influence. They're precisely wrong instead of approximately right. Adding rules doesn't fix a method that can't see cause; causal vs rule-based vs data-driven attribution in GA4 compares the three approaches side by side.

Beyond Correlation: The Shift to Causal Inference

Causal inference is the shift from correlation to causation: a method that estimates the business impact of your marketing. Unlike attribution, which only tracks events, causal inference uses counterfactual analysis to estimate the incremental sales each channel generated. It tells you what would have happened if an ad had never been shown, which is the value of the spend.

The way out of the cycle of flawed models is to shift from a correlational mindset to a causal one. Stop asking, "What channels did a customer touch before converting?" and start asking, "Would this customer have converted without seeing this ad?"

This is the central question of causal inference. Instead of just observing a sequence of events, it estimates what would have happened in a world where a specific marketing action wasn't taken. The difference between what happened and that counterfactual is the incremental impact of your marketing. The simplest way to get it is an experiment: hold spend back in one region and compare. Causal attribution estimates the same thing from data you already have.

Understanding Paths to Purchase

Customers rarely move through a neat funnel. The path from awareness to purchase is a web of interactions in which one touchpoint sets off behavior that ends in a sale days or weeks later. For illustration: a TikTok video that gets few clicks can still lift branded search and direct visits over the following two weeks. A click-based model credits that lift to branded search and Direct; pausing TikTok in one region shows where it really came from.

Measuring cause also shows the opposite problem: channels that take credit for sales that would have happened anyway. Paying for branded search ads is often a case of paying for customers who were already typing your name. Other channels work the other way round: a podcast sponsorship can sell little on the day and still lift how well your Meta ads convert weeks later, which only a test over time will show. Why correlation-based marketing decisions waste budget walks through the arithmetic.

The Power of Knowing Why: From Guesswork to Growth

Knowing why customers buy changes what you can defend in a budget meeting. Platform ROAS tells you what each platform claimed; incremental sales tell you what the spend added. With the second number you can cut waste and argue for the channels that earn their budget.

It changes the conversation too. When you know why customers buy, you stop defending last month's dashboard and start deciding where the next euro goes.

From Wasted Spend to Precision Investment

For illustration: a Dutch beauty brand splits its budget across Google Ads, Meta Ads and TikTok, and its attribution platform reports a healthy 4x ROAS on all three. Revenue is flat. A regional holdout on each channel shows that Meta drives most of the incremental sales, Google Ads mostly captures demand that already existed, and TikTok takes sales that organic search would have brought in anyway.

With that, you can move budget: cut Google spend where it only captures existing demand, put the savings into the Meta campaigns that proved themselves, and test new TikTok creative before you scale it. Same total spend, more of it working. The attribution ROI calculator puts a number on what better attribution is worth to you.

What It Means for the Modern Marketer

Your job changes with the method. Instead of pulling levers and reporting ROAS, you walk into a meeting with your CFO and show what the marketing added to revenue, with a test behind the number. That's how you earn trust and the budget to scale.

Your First Step Towards Causal Marketing

Taking your first step towards causal marketing means questioning the numbers you have and testing before you trust them. You don't need a PhD in statistics, a data team or new software to start.

Three steps, in this order:

1. Question Everything: When Meta reports a 10x ROAS, ask how many of those customers would have bought anyway. Then do the arithmetic: add up the purchases Meta, Google and TikTok each claimed last week and set the total next to your Shopify orders for the same week. The difference is double counting.
2. Rebuild Last-Click Yourself: Export revenue by session default channel group from GA4, leave Direct out, and compare each channel's share with what its platform claims. The channels where the two disagree most are the ones to test.
3. Embrace Experimentation: Run a geo test: keep a campaign running in one province of the Netherlands and pause it in a comparable one. The difference in sales between the two is a rough measure of incremental lift, and it's the foundation of causal thinking. Confirming a weak channel with a holdout test walks through the set-up.

Rule-based attribution isn't coming back; GA4 has already dropped most of it. What replaces it is less tracking and more testing.

Frequently Asked Questions

What is the main problem with marketing attribution?

It measures presence, not effect. Attribution records which touchpoints appeared before a purchase and splits the credit by a rule, without checking whether any of them changed the outcome. That's how budget ends up on channels that were only present, and why platform ROI and real ROI drift apart.

How is causal inference different from marketing attribution?

Attribution splits credit; causal inference estimates effect. Attribution assigns each sale to touchpoints by a rule. Causal inference asks whether the sale would have happened without a given channel, and estimates the incremental lift from the difference.

Is causal inference difficult to implement?

Not the first version. A regional holdout needs a spreadsheet, not a statistician: pause a channel in one region for two weeks and compare revenue with the rest. Rebuilding last-click from your GA4 export takes an afternoon. Later, if you want the same question answered for every channel at once, a causal attribution tool like Causality Engine reads it from that same GA4 export and puts last-click beside it.

Why are traditional attribution models no longer effective?

Because they can't see the whole path any more, and they never could see cause. Consent rules such as GDPR, browsers that block third-party cookies, and platforms like Meta and Google that keep their data inside their own walls leave gaps in every click path. GA4 has already retired first click, linear, time decay and position-based models, and its data-driven model is trained and reported by Google, which also sells you the ads.