First click, last click and linear multitouch attribution models, part 3

In the previous article, I described what is attribution and how the default, last click model works. The problem with this model is that it doesn’t give any credit to the channels that were driving value at higher stages of the funnel. This model is not having in account those first channels that attracted the attention, generated awareness and drove the user to consider purchasing the product. The last click model or default model will only give credit to the last converting channels.

To understand which have been the channels that had the first interaction with the user we can use the first click or touch interaction models. This model gives credit to the channels on top of the funnel. The problem with this model is that it won’t help us make any decisions by using it in isolation. This model by itself doesn’t give us enough insights on what channels will be generating the conversions and revenue.

If we look at the last click or first click models in isolation, these won’t tell us the truth or the effectiveness of the marketing activity. These are just showing part of the picture. The best thing we can do would be to look at both models, as they will help us tell the story of what channels contributed at different stages.

To put an example we can use Google Analytics to visualize these models, this has some easy to configure reports that we can use. To access them go to your Google Analytics account, choose from the left-hand side menu Conversions, Attribution, model comparison tool. Then select from the drop down, Conversions and value.

This report gives us the ability to model the data depending on the model we choose to apply. Looking at the example below we can select from the drop down to compare the first click vs the last click models, the model will attribute value to the channels that sit at each of these extremes. This will help us categorize where these channels sit:

First click – top of the funnel

Last click – bottom of the funnel.

What we can see here are values that represent the values in transactions and revenue for each channel. If the channel has a high value for the first click, this means it was successful at interacting with users at the top of the funnel. If the channel has a higher value at the last click interaction column, this means it was good at driving conversions and sales or bottom of the funnel.

Looking at the data that we have, we could categorize the activity of each channel in:

1. High in the funnel channels or first click interactions, these are opener channels and usually are:

• Paid search generic terms
• SEO – generic terms
• Display prospecting

2. Bottom of the funnel channels or last click interactions, these are closer channels

• Direct
  
• Email

As can be seen, by the figures in the table, there’s no straight line that divides which channels as sitting on top or at the bottom of the funnel. But looking at the data with these models gives us an idea of where these channels have a higher influence.

We can also sophisticate the reports by improving the categorization of channels and subcutáneos. For example; In this report, we don’t have branded paid search classified as a channel, but if we had, this would be a good example of a last click interaction channel that would give us a more clear view of where search channels index and the difference between branded search and non-branded search. As brand searches are far down the funnel compared to the more generic searches that are higher in the funnel.

As a strategic decision, we could use generic non-brand paid search as a way to drive users to the top of the funnel to generate awareness and consideration.

If our goal is to understand how each channel influenced the user journey why are we liming ourselves to first and last click? Why not look at a model that will show us the whole journey? Here is where linear models come in.

This is a model that distributes the value of the conversion evenly across all the channels that interacted with the user.

We can go back to the example in Google Analytics and add to the comparison model table in the drop-down menu and from here select linear. We can then sort the data in the table by conversions. By doing this we will be able to sort and visualize what channels are indexing more as a middle of the funnel channels. In this case, we can see that Direct, paid search and organic search are indexing heavily here. This means that the user was interacting with these channels across the whole length of the funnel.

Now that we have done an overview of this model and see how it’s distributing the credit of a conversion. Let’s think if it makes sense to use it as a model to make marketing decisions.

If we want to have an effective marketing strategy, in reality, this model won’t be the most accurate in giving us a picture on the impact of each channel at the different stages of the funnel. Be aware that by nature some channels are good at generating awareness and driving consideration while others are good at driving conversions. We need to look at the separate models (first and last click) to understand how this is happening, unfortunately, the linear model is not great at providing this information.

If you’re looking to start applying these models you might want to structure the architecture of a data driven acquisition model framework first.

There are other advanced models that can provide a unified and consistent measure of the value of each channel, I will talk more about them in my next blog post, stay tuned.