One of the main assets of digital marketing is the ability to measure interactions. This allows you to track the number of conversions generated by each distribution channel (SEM, Programmatic, Social Networks, Email…) in order to identify the most profitable ones and thus determine the best budget allocation.
This is where the notion of attribution comes into play, allowing you to distribute the conversions recorded between the marketing levers according to a relevant model.
Let’s take a concrete example:
If a customer buys from my site after clicking on an SEM ad, should we consider that the sale was made only because of this last click? Should we attribute 100% of the revenue to this ad alone?
In reality, it is not so simple, because before buying, the customer may have learned about the brand through a video on their Instagram feed; then been exposed to a programmatic display banner; compared different products on Google Shopping before finalizing their purchase after clicking on a text ad.
These different interactions are essential contact points to be present throughout the customer’s purchasing process, both in the awareness, consideration and purchase phases.
The choice of the attribution model is then crucial to know which are the effective levers to generate conversions and calculate the profitability of its investments.
All tools using attribution models will then split each transaction and its revenue across multiple acquisition levers.
A conversion that generated $100 in revenue might be split $30 to Instagram, $20 to SEM and $50 to SEO.
What are the different attribution models?
- The last click model attributes the conversion to the last lever that generated an interaction such as a click, for example. This is the default attribution model of Google Analytics or Google Ads and therefore the most commonly used, but not the most accurate one, as it favors end-of-funnel levers like search and retargeting.
- The first-click model, which is more rarely used, attributes 100% of the conversion to the lever that generated the first contact with the buyer. The limitation of this model is that the more interactions there are before a transaction, the more complicated it is to know what the first interaction was and if it was really decisive in the process.
- Time decay gives more weight to channels that generated an interaction shortly before the purchase. On the opposite, the longer an interaction was made before the transaction, the less it will be valued.
- The linear model assigns the same share to each lever that participated in a conversion.
- The position-based attribution values the first and last point of contact with the customer.
- The data-driven attribution is the most advanced and reliable model, as it assigns a greater or lesser share of revenue to levers based on their contribution, based on the analysis of conversion paths. This is the Shapley value, based on game theory.
Ideally and when possible, it is better to use the data-driven model. Google Ads will make it its default model by early 2022.
However, while this model is the least imperfect, it is not without limitations. Limitations that it is important to keep in mind to avoid analysis bias:
Impressions are not always taken into account as interactions on certain tools, which will favor click-generating levers like search, to the detriment of levers further up the tunnel like social networks or programmatic. Levers that are nevertheless essential to the buying process.
Even on attribution tools that take impressions into account, the installation of impression pixels is not possible on social networks. Only a few solutions such as the Wizaly attribution tool allow to extrapolate the volume of impressions.
Data-driven marketing requires a high volume of conversions to work.
As customer journeys become more complex, attribution becomes an essential concept to consider. While no perfect model exists, the most reliable is data driven. This model can be used on Google platforms, but also for attribution or web analytics tools like :