Attribution Modeling: A Complete Guide

Add up the conversions claimed by Meta, Google Ads, Klaviyo, and GA4, and the result can exceed actual revenue by 30–80%. You only need to see it once in a report to know something does not add up.

None of these systems is deliberately lying. Each simply views the world through a different lens. Understanding why those views differ is the foundation of attribution work.

Avinash Kaushik puts it plainly: “The only use for last click attribution now is to get you fired.”

What attribution is and why it matters

Attribution is the process of assigning a conversion to a traffic source. Every conversion in every system IS attributed. The question is not whether you measure it, but HOW you measure it.

Marek Kobulský from Ecommerce-academy.cz uses a helpful analogy: a weather forecast is a mathematical model. There is only one actual weather outcome, but there are several models for predicting it, and each produces a slightly different result. Attribution works the same way. There is one conversion, but several models for assigning it.

In our “interlocking gears” view of marketing, this means one thing: channels cooperate rather than compete. Attribution is a tool for understanding what that cooperation looks like.

The customer journey and why every system tells a different story

Imagine a typical journey. A customer sees your Facebook ad. A day later, they return through organic Google search. Then they see a remarketing ad on Seznam, open an email from Klaviyo, and finally buy through direct traffic. Who “won”?

Every system sees only its own piece:

• Meta: “They saw an ad and bought one day later—my conversion.” That includes view-through attribution, which GA4 does not see at all.
• Email (Klaviyo): A five-day window after an email open. That is generous attribution, and Apple Mail Privacy Protection generates false opens, inflating the numbers further.
• Google Ads: Brand and Shopping ads sit near the bottom of the funnel. The conversion is close to the click, so the discrepancy tends to be smaller.
• GA4: A cross-channel view with a data-driven model. It is the closest thing to a neutral observer, but it is still only a model.

Upper-funnel awareness is even harder because it works with impressions rather than clicks. It is more difficult to measure and assign a value to, but that does not mean it has no value.

Attribution models and what changed

If you encountered attribution years ago, you may remember the list: first click, last click, linear, time decay, position based, and data driven. Each distributed credit differently. The landscape has since changed.

In September 2023, Google removed linear, first-click, time-decay, and position-based models from GA4. Data-driven attribution became the default, with last click as the fallback. Universal Analytics had used last non-direct click by default, and many people still interpret data through that older mental model without realizing it.

Kaushik offers one of my favorite analogies: first-click attribution is like giving your first girlfriend 100% of the credit for the fact that you married your wife.

The important insight is that there is no “best” model. What matters is understanding what attribution does, accepting its imperfections, and following one primary metric as a trend. You can then compare models in greater depth, accounting for the context in which each system operates.

How I approach attribution in practice

I follow one primary metric as a trend, typically revenue or ROAS in GA4. When I go deeper, I compare attribution models across systems in the context in which those systems operate. Meta sees view-through activity, GA4 sees cross-channel behavior, and Klaviyo sees opened emails. Each perspective has value.

Channel grouping is foundational. Attribution is meaningless without thoughtful source segmentation. You need to know what you are measuring before the results can tell you anything. That segmentation must also connect to the business plan and KPIs. Are you optimizing for volume or profit?

Instead of looking for a single truth, I use triangulation: platform data, GA4, and incrementality tests. For more advanced clients, I add a post-purchase survey. Each source tells a different story, but the overlap among those perspectives brings you closest to reality.

Rand Fishkin warns about an important consequence: attribution obsession drives overinvestment in channels that are easy to measure at the expense of brand building. If you make decisions only from what can be measured directly, you will naturally underinvest in everything that builds awareness. That is dangerous over the long term.

---

Attribution is not about finding the truth. It is about understanding how channels complement one another and accepting that perfect measurement does not exist.

“Let’s measure it”—while remembering that the numbers are a model, not reality. A model that helps you make better decisions is still better than no model. The question is not, “Which channel won?” It is, “How did the channels work together so the entire business could win?”