Attribution Models, Explained With One Order
First click, last click, linear, data-driven, order-verified.
My first hire, a media buyer called Jonas, spent his first week in a war with our Google agency. He said brand search was stealing credit. They said Meta was a money pit. Both of them were reading the same orders through different models. Neither of them knew that was the argument.
The one-sentence version
An attribution model is the rule that splits one order's credit across the ads that touched it. The order does not change. The rule does. Pick a different rule and a different campaign looks like the winner.
The setup
Three orders from one week
Order #1041: a Meta prospecting ad, then a Google brand search, then the email that closed it. Order #1042: prospecting, then retargeting. Order #1043: someone typed the brand into Google and bought. $290 in total. Now split it.
Try it: three orders, four channels, one week
Same week. Pick a model. Watch the winners change.
| Order | Journey | Value |
|---|---|---|
| #1041 | Meta prospecting → Google brand → Email | $90 |
| #1042 | Meta prospecting → Meta retargeting | $140 |
| #1043 | Google brand | $60 |
Meta prospecting
$0
Meta retargeting
$140
Google brand
$60
$90
Last click: total credited $290. It always adds up here.
Email and brand search look like heroes. Prospecting, which started two of three orders, gets $0. The total never changes. Who gets it changes completely.
A model never changes how much you sold. It only decides who looks like a winner.
The models
Six rules, one sentence each
Last click
The final touch takes 100%. Rewards whatever is closest to the checkout: brand search, email, retargeting.
First click
The first touch takes 100%. Rewards discovery, ignores everything that closed the sale.
Linear
Equal shares. Treats a brand search like an Instagram reel.
Time decay
More credit the closer to the purchase. Last click in a nicer suit.
Data-driven (platform)
Weights learned from paths, but only paths the platform can see. Meta's version cannot see Google.
Order-verified
Starts from the Shopify order, sees every channel, weights learned from your journeys plus surveys. Cannot exceed the order.
Why it matters
Every model has a favourite campaign
Last click loves retargeting and brand search. First click loves prospecting. Linear loves whoever shows up most often. Whichever model your dashboard defaults to, that is the campaign you will over-fund.
Jonas and the agency eventually agreed on one thing: they were both right inside their own model, and both wrong about the business. That is the normal outcome.
The fix
Pick the rule that cannot flatter anyone
Two tests for a model you can budget on. One: it sees every channel, including email and the survey answer. Two: its totals equal your Shopify revenue. Platform-native models fail the first test. Summing platform reports fails the second.
TrueROAS uses order-verified attribution: the Shopify order is the unit, every touch is recorded server-side, the survey fills the view-only gaps, and a model trained on your store's journeys decides the split. You can still flip to last click or first click inside the dashboard when you want to see what a platform sees.
Honest footnote
When last click is fine
If 90% of your orders have one touch, every model gives the same answer and this page is academic. Check your journey length first. Most stores over $20k a month are surprised.
Questions people actually ask
Straight answers
Which model does Meta use?
Meta reports its own touches only, with a 7-day click and 1-day view window by default. Inside that, it is effectively last touch among Meta ads. It does not know about Google or email at all.
Which model does Google Ads use?
Data-driven attribution by default since 2023, but only across Google's own touches. A Meta click is invisible to it, so a brand search still gets full credit.
What is data-driven attribution?
A model that learns weights from your conversion paths instead of a fixed rule. Good idea. The catch is that the platform versions only see their own paths, so they learn from a partial picture.
So which one should I use?
For allocating budget: a model that sees all channels and starts from your orders. For judging creative inside one platform: that platform's own view is fine. Do not mix the two.
See your own numbers, verified against your orders.
Free under $10k/month in revenue. 14-day trial above that. Nothing changes in your ad accounts. You will know within a week which platform has been telling stories.
"I built this because I was making $50k-a-month decisions on numbers I knew were wrong. If TrueROAS does not show you something Meta or Google got wrong in the first two weeks, tell me and I will help you cancel."
Rasmus Arvidsson
Founder, TrueROAS. Answers support himself.
Keep going