Tracking is broken. Your direct traffic isn’t really your direct traffic, says Russell from Cubed. There is an inherent problem.
The attribution models Google gives us are incorrect. The future will be data-driven, as Rebekah said.
So… we live in a world of multi-channel, and we need a better understanding of what each visit is and where it comes from. The way the attribution modelling works in Google is only based on channels, so will only inform optimisations at a channel level.
The main thing we want to do, though, is to understand how to drive ROI. GA’s last click model is definitely wrong, so if we put our data through proper attribution models, we can better understand and associate costs to our visits too. There is always a need for ROI in any channel and we need to attribute ROI to every channel that plays a part.
So how do we do this?
The kind of system we have within Google is at a channel level. The reason we have to use machine learning is that for PPC, we need a deeper level of understanding. If you’re running stuff through AdWords or Doubleclick, you’re trusting Google’s data – and with the free attribution model it will release later today, we’re having to trust again.
The cool thing with machine learning is that it will understand the links between channels. Those people who have done four or five things with us but haven’t converted, we can work out what the chances of conversion are if we do anything else.
We can use Russell’s tool to work out the probability of conversion, to be able to amend our bids accordingly.
Russell says to check out this article from Seer, where they look at the direct traffic issue.
Direct is a clear indicator of brand awareness. But if we have it all clogged up with untagged PPC or untagged offline campaigns etc etc, we’ll never know its value.
Attribution is so important, but don’t always believe what you see.