Attribution Analysis for Clickstream Data
At a Glance...
Marketing attribution is essential for businesses to help assess the value or ROI of a given channel. So what models should be followed? Here, Analytics Manager, Maria, outlines a twofold strategy for clickstream data that’s sure to provide insightful results.
What type of attribution model are you currently using for your conversions? What type of analysis are you running to understand users’ behavior? Does visit 1 or visit 2 or visit X have an impact in causing a conversion event? If these types of questions have been raised during your performance meetings, we are here to tell you why you shouldn’t limit your attribution models to first / last touch, or simply rely on the multitouch models that are available in your tracking tool.
In this article we are going to cover two types of analysis which can be done on clickstream data that will help your business understand the weight of each of your channels when it comes to completing a conversion.
(Please note that our analysis is being executed with Python libraries)
Is it possible to estimate the likelihood of a user to perform an action based on their previous touchpoints? Yes, this can be achieved by using a Markov chain – a statistical model that considers user behaviour and calculates the probability of each event for happening based on previous touchpoints.
The next two examples will illustrate how you can apply this model to your data:
1) Likelihood of a user coming back to your site
After loading the data to our python script, we can display how likely is for someone to come back to the site depending on their last touchpoint. The thick arrows in the image below show the most likely transition from this channel to the next channel. The second edge, depicted by a thinner arrow, displays the second most likely transition from this channel.
This type of visualisation helps to identify correlations between channels that are not always easy to spot when only looking at data in conventional structures.
2) Markov chain attribution modelling
A Markov analysis can help you understand how much more likely someone is to visit and convert if they have previously visited X. By using clickstream data, the model outlies the probability associated with a sequence of events occurring based on the previous event(s).
The Markov chain attribution modeling is based on the analysis of how the removal of a given touchpoint from a defined user journey affects the likelihood of conversion. In summary, this attribution model returns a removal effect score for each touchpoint by running simulations of what would have happened if a specific channel is not present in the customer’s journey.
Think about the following example, in which two users’ first touchpoints are paid social, where they then take two different journeys:
What would happen if we removed paid social? The simple answer is that by removing the first channel we probably wouldn’t have reached any of the users and the probability of them converting is zero.
But if we remove direct instead, the probability is now 0.50, as one of the users might still complete the journey:
Lastly, removing paid search has a very similar effect to removing Paid Social. The probability of any user to convert goes back to being zero:
Since paid social and paid search have the same removal effect, we can see how valuable these two channels are in our example. Now imagine how powerful this model can be if you feed it with enough data. The learnings from executing this type of analysis can go from budget allocation to taxonomy improvement and more.
Clickstream Path Analysis
There are a lot of useful tools out there that can help you visualize user paths and identify the most frequent paths to enable you to see which pair of channels have a higher, stronger correlation between them. If you are considering turning one channel off, you should review (together with the Markov analysis) how important that channel is to your user path. Understanding the role each channel plays in your customer journey is essential when building your marketing strategy.
The customer journey can be displayed in the following way by using python packages:
In conclusion, by combining python packages and marketing knowledge you can take your planning and strategy to the next level. If you are interested in getting started with python or have questions about the described model do get in touch with us and we’ll be happy to help.
Maria Rivelles, Performance and Analytics Manager