The Ecomm Analyst

Short version first. Use last click as your default reporting model, keep one other model available as a cross check, and stop expecting either one to tell you what caused the sale. Attribution models are accounting rules, not measurement. They split credit across the touchpoints you managed to observe, and they say nothing about what would have happened if you had turned the channel off.

That distinction matters more than the model you pick. I have watched brands spend a month arguing about time decay versus position based while their UTM parameters were broken on half their email sends. The model is the last thing to fix, not the first.

What an attribution model actually is

A customer touches your brand several times before buying. Maybe a Meta ad, then a Google search for your brand name, then an email three days later. Your store records one order. The attribution model decides how to divide that one order across the three touches.

Every model is answering the same question, which is how to split one conversion across several touchpoints. None of them answer the question you actually care about, which is whether any of those touches was necessary. That question needs a holdout test, not a model.

The six models you will run into

Last click

The final touch before the order gets everything. This is what Shopify reports by default and what most operators are looking at whether they know it or not. It systematically overpays branded search, retargeting, and email, because those tend to be the last thing a customer touches on the way to a purchase they were already going to make.

First click

The first recorded touch gets everything. It is most useful as a mirror of last click. If a channel looks strong on first click and weak on last click, it is doing discovery work. I went through running these two side by side in first touch vs last touch without overcomplicating it.

Linear

Every touch gets an equal share. Three touches, a third each. It is easy to explain and it has no opinion, which is either its best feature or its worst depending on your view. In practice it flattens everything and makes it harder to see which channels are doing something distinctive.

Time decay

Touches closer to the purchase get more credit, on a decay curve. It is last click with the edges sanded off. If you find last click too harsh but you do not want to pretend an impression from three weeks ago mattered, this is the compromise.

Position based

Also called U shaped. The first and last touches take the bulk of the credit, usually 40 percent each, and everything in the middle splits the remaining 20. The logic is that discovery and closing are the two moments that matter. That is a defensible assumption, and it is still an assumption.

Data driven

The platform builds a model from your own conversion paths and assigns fractional credit based on which paths converted. Google and Meta both run a version of this. It sounds like the answer to everything. In practice you cannot audit it, you cannot reproduce it, and the vendor grading the homework also sells you the media.

So which one should you use

Pick last click as the number you report on. It is conservative, it is stable, and everyone on your team already understands it. Then pick one contrast model, either first click or position based, and look at the two together once a week. The gap between them is the interesting part. A channel that only shows up on last click is closing demand somebody else created. A channel that only shows up on first click is creating demand somebody else closes.

What I do not do is switch reporting models when the numbers get uncomfortable. If Meta looks bad on last click and you move to position based to make it look better, you have not learned anything about Meta. You have learned something about yourself.

For the actual mechanics I use ThoughtMetric, which sponsors this blog. It pulls order level data straight from Shopify and lets me look at the same period under different models without rebuilding a report each time, which is the only reason I bother comparing models at all. Northbeam and Triple Whale do a version of the same thing at higher price points. If budget is the constraint, GA4 will show you several models side by side, though its order totals will not match what Shopify recorded.

Common questions

Is multi touch attribution better than last click?

Not more accurate, just different. It spreads credit across observed touches, cannot see the touches it never recorded, and still cannot tell you whether a channel was necessary to the sale.

Which attribution model do the ad platforms use?

Each one uses its own, applied only to its own data. That is why their reported numbers overlap and add up to more revenue than you actually made.

Do I need to pick one model and stay with it?

You need one for reporting so the number stays comparable week to week. Looking at others alongside it is fine, and it is where most of the useful signal lives.