CXL’s Growth Marketing Minidegree Review — Attribution [Week 7]
This is my seventh week with CXL, Learning Growth Marketing Minidegree, and in this week, I am going to be talking about an important topic, which is Attribution.
Intro
Between SEA, retargeting, SEO, Social Media, display and the many existing Adtech tools, “attributing” the conversion, that is, determining which channel was decisive in the purchase decision, is a real challenge for marketers. An attempt at a definition and a brief overview of the different attribution models.
According to Wikipedia, Attribution is “the process of identifying a set of user actions (“events” or “touchpoints”) that contribute in some manner to a desired outcome, and then assigning a value to each of these events”. Clearly, it is about determining the most decisive channels in the buying journey in order to know where to invest first. Is it exposure to a banner, emailing, SEO?
The stakes are therefore high and the award has grown over the years as a discipline in its own right. So far, despite the many attribution models that exist, there is still no silver bullet.
A Quick Run-through of Some Common Attribution Models
There are several marketing attribution models, some traditional, others newer. Each has its advantages and disadvantages. You need to choose the one or those that best meet your business needs. Let’s go through the most important ones.
1. Last Click Attribution
The “last click” model is the most traditional, the one used by default by most web analytics tools — starting with Google Analytics. As the name suggests, this model consists of attributing the conversion to the last click recorded. This is a very basic model, and therefore largely insufficient, since a conversion is always preceded by several interactions. For example: a user discovers your product on Twitter. He types the name of your site in his Google, clicks on your site link and buys the product (very basic conversion scheme, we agree). Following the last click attribution model, 100% conversion credit will be attributed to “Google / organic” traffic, to the detriment of other conversion levers that have intervened (Twitter in this case).
This simplistic attribution model generally overestimates the role of AdWords (and more generally SEA) and SEO levers. The last click model does not (or rarely) reflect reality. In reality, it is almost always more than one that leads to conversion. At a time when customer journeys become ever more complex, the last-click model is becoming increasingly unsuitable.
2. Last Non-Direct Click Attribution Model.
This model is similar to the previous one. With one difference: it does not take direct traffic into account. If a user discovered your offer on Twitter, then did a Google search for more information, then came back to the site two days later by typing the URL of your site in their browser, the last click taken into account will be that of organic traffic.
This model has one advantage: it is simple. Excluding direct access is justified in most cases. A user who types the site’s URL into his browser learned about the site through other levers. So it is most of the time absurd to give the credit of a conversion to direct traffic. But, at the same time, this model retains the drawbacks of the classic last click model. Ultimately, this is just a variation of it.
3. First Click Attribution
This model, along with the last click model, is one of the two most used. It’s a traditional model of giving full credit for the conversion to the first point of contact / channel / click. In other words: it is the opposite of the last click model. This model has the advantage of simplicity, but retains the disadvantages of the previous models reviewed. The Mazeberry site uses a very telling metaphor in one of its articles: it amounts to attributing your success in the bac to your passage in second class. We could add: that amounts to attributing to his first love of youth all the merit of his marriage. In fact, this attribution model is much less common than the last click one.
4. Linear Attribution
All of the models listed above have one huge flaw: they give all the credit for a single channel conversion. The reality is denied: every conversion, except in exceptional cases, involves several channels. The models that we are going to present to you now have their flaws, but also a great merit: they spread the credit of the conversion over several channels / touch points / clicks. Let’s start with the linear model. In this model, the merit of the conversion is distributed equally among the different channels that intervened in the conversion process. The downside of this model is that it denies the respective contribution of different channels. In any given conversion, there is always one channel that played a bigger role than others. The linear model is therefore an “ideal” model.
5. Time Decay Attribution
This marketing attribution model takes into account the position of marketing channels in the conversion process. According to this model, it is the touchpoints / channels closest to the conversion that get the most merit in the conversion. The logic behind this choice of attribution is that the most upstream channels by definition failed to lead to a conversion.
Note that there is an inverse (“descending”) model, which gives the most merit to the channels located further upstream of the conversion.
6. Position-based Attribution
In the latter model, the first and last channel of the conversion path are given the most weight. We speak of a parabolic model, or sometimes of a “U” model. The first and last channel each receive 40% of the conversion credit. The intermediate channels share the remaining 20% equally. You are free to vary this distribution (30/40/30 instead of 40/20/40 for example). The parabolic model is hybrid, in that it combines the increasing model and the decreasing model.
Custom Attribution Modeling
The models presented above are standard models, mainly having a heuristic function. They each respond to a problem, but the relevance of each of these models can be questioned. The ideal is to build your own attribution models in addition to those offered by Google Analytics. You can also use the R language to create more complex attribution models.
However, keep in mind that your model can never be perfect anyway. An attribution model is always necessarily based on more or less arbitrary assumptions about how your customers’ journey works and on an element of instinct.
To create a personalized attribution model in Google Analytics, go to the menu Conversions> Attribution> Create a personalized model (scroll bar). This allows you, from a reference model, to adapt and personalize the rules.
This was all for this week, and next week, I will be talking about landing Page optimisation and Product Messaging.
A noter qu’il existe un modèle inverse (“décroissant”), qui attribue le plus de mérite aux canaux situés les plus en amont de la conversion.
