On June 10th 2016, Wolfgang Digital held their annual marketing event in The Foundry, at Google in Dublin. David and Zoran discussed Channel Attribution during part two of the event.
38. MachineLearning(AI)
• A type of AI thatgives computers abilityto learn
withoutbeing explicitly programmed
• They willgrow and changebased on new data
• Facebook News Feed
42. KeyTakeaways:
• Don’t look at channels in isolation
• Understand their placein the funnel
• Give them creditthey deserve
Editor's Notes
Thank you David.
Now that you cleaned your data and defined your channels in a meaningful way it’s time to start thinking about giving them credit they deserve.
So, how many of you heard about attribution? How many of you work on it?
Thank you David.
Now that you cleaned your data and defined your channels in a meaningful way it’s time to start thinking about giving them credit they deserve.
So, how many of you heard about attribution? How many of you work on it?
Thank you David.
Now that you cleaned your data and defined your channels in a meaningful way it’s time to start thinking about giving them credit they deserve.
So, how many of you heard about attribution? How many of you work on it?
To crack this problem we have to understand default attribution models.
First one is last click attribution. All credit goes to the last channel before a conversion. It’s like giving the last speaker today all the credit for a great conference. So, Alan gets all the credit. Good for you Alan.
The next one is the last non direct click. It’s the same as the previous one, but we’re just ignoring direct channel. This last icon represents the direct traffic channel.
Then we have the first click model. It’s the opposite of the last interaction model. The first interaction gets all the credit.
Linear attribution model is next. It credits all channels equally. But we know they are not equal and this is therefore a bad model. We will only give 2 prizes after the quiz later on. You can’t all be winners. That should apply to attribution as well.
Next one is time decay attribution. Channels are receiving more credit the closer they are to the conversion. That’s much better with some logic built into it.
And finally, position based attribution. First and the last click get more, while the middle interactions share what’s left. This one too has some logic to it.
However. They are just assumptions about your business and your customers.
In an ideal world, you’d create a custom attribution model.
We’re gonna zoom-in and explain what you can do with this.
Maybe you want to use a position based mode, but assign percentages differently. So you turn this into this.
Maybe you want to change the lookback window to what makes sense for your business. Recently, a client asked what would happen if we change our conversion window from 30 days to 3 days in AdWords.
Moving on. Maybe we want those middle interactions to get credit base on how many pageviews they received. So if organic search received 3, adwords 4 and referral only 2 pageviews, this model goes from this to this.
And finally, if you don’t want bounced visits to receive any credit, no matter the channel, you can turn this model into this model. There’s lots of other ways you can apply custom credits on this screen.
So from the first dimension, a channel list, you understand that multiple channels are driving traffic to your site. From the second dimension, or conversion paths, we understand that channels don’t work in isolation, but they are supporting each other. And finally, when we take engagement metrics like bounce rate, pages per session, their position in the path, lookback window, revenue and everything else, we get to the attribution model and can then understand how much credit they deserve. This is a 3d view of your channels you need to optimise your business online.
It is kind of complicated and it takes a lot of effort. But Google has all this powerful technology and they can do it for us. And fast.
So, how Google uses this for attribution?
It’s called data-driven attribution. Google looks at all your conversion paths and calculates how likely a conversion is with every mix and sequence of channels. We’re effectively following this trend of leaving the hard work to a computer or a robot and its artificial intelligence.
On the screenshot I’m about to show, you’ll be able to see a custom model generated by Google. Darker colors mean more credit should go to those channels in certain positions.
We might see something like this. So for this client, Google gave us a model that resembles a time decay model with later channels receiving more credit. You see darker colours are closer to the conversion.
Similarly, Google introduced this in AdWords. We can now go even deeper looking at how different adwords campaigns interact with each other. We know a lot of brand campaigns convert because people were previously informed by one of the generic campaigns. Now we can give them credit they deserve.
So getting a good understanding of how your business works online requires three things: don’t look at your channels in isolation. They are not independent of each other, they work together. 2nd, understand where your channels are in the funnel, whether they’re assisting or tend to be the last one before a conversion. And finally, give them credit they deserve. Find that right attribution model.
And don’t be afraid to use artificial intelligence. It might look scary, but it’s actually our friend. Thank you