Your Attribution Model is all wrong

“Yeah we’re doing attribution, oh yeah we have a probabilistic approach, We are able to bid on the third keyword that generated incremental conversion in the path before a direct purchase”.
How many times I’ve heard Performance guys telling those bullshit. Unless you’re able to afford a 15–25K/month software attribution (VIQ or Neustar MarketShare) your attribution for first time buyers will be screwed.
Why?
1)Cross-device
2) Cross-device
3) Cross-device
#For users not logged in your website (no User_ID)
First you need to understand that Google Analytics is not a user-centric view. It’s a device-centric view, it’s neither a deterministic nor a probablistic approach to cross-device.
You are using the data-driven attribution in GA 360, WOW! What A fancy word for a such monster shit.

What is the point of making attribution if you don’t see the correct path?
If you have user A that has 5 interactions to online media in 3 different touchpoints/devices:

In GA you should see:
UserA {
touchpoints <-c(“Desktop”, “Mobile”, “Tablet”)
media_exposition <- c(2,1,2)
}
What you will see instead:
UserA & UserB & UserC {
touchpoints <-c(“Desktop”, “Mobile”, “Tablet”)
media_exposition <- c(2,1,2)
}
Google starts computing the shapley value probability conversion of an incorrect path missing 2 touchpoints.
You don’t need to be a Rocket scientist to understand that the method used is completely wrong. Take decision on your performance budget based on those insights is SCIENCE FICTION. 😂
Oh, you’re doing attribution with BigQuery you’re sending hits instead. The same science fiction, unless during the session you force the user to login to your web/app which will be great for UX&CRO no doubts.
Same goes for Kissmetrics, Mix Panel or Woopra they’re all cookies based GUI solution, forget about attribution.
By the way we just decided to move things forward with Neustar MarketShare in order to build models that integrates the post-view & cross-device with an advanced discrete choice models that computes off-line factors.
I’ll keep you posted on the efficiency of the model.
Cheers,
Gus.
