Multi-Touch Attribution: Why, What, How?
Multi-Touch attribution (MTA) is a topic widely discussed in the field of e-commerce and marketing, yet the number of companies actually carrying out conversion attribution using MTA remains low. There are various reasons for this: the marketing department bases its results on the last-click model or the required resources to switch to MTA are simply not available. This blog will discuss why conversion attribution via the MTA model is important, what a good MTA algorithm looks like and how it can be applied.
Why is Multi-touch attribution relevant?
One of the best analogies to highlight the importance of MTA is drawn between a bank robbery and conversion attribution. Robbing a bank requires a lot of work to be carried out by multiple robbers; without a planner, the robbery will not take place, without brute force the money will not be handed over and without someone driving the getaway car, the robbers cannot escape. It would be odd to then hand over the entire loot to the driver, simply because he was the last robber to perform a task during the robbery. This is, in essence, what the last-click model does to touchpoints in a customer journey. The attributed value, however, does not mean the touchpoint really was responsible for the conversion. The other two robbers were, after all, instrumental to the robbery, but receive no credit. Results obtained using the last-click method are therefore a misrepresentation of the path to a conversion. A good attribution algorithm does assign to each touchpoint the value it contributed to the conversion.
What does a good attribution algorithm look like?
For an MTA algorithm to be considered a ‘good’ algorithm, it needs to have the following four properties: fairness, data driven, interpretability and economic efficiency. An algorithm for which these four conditions hold is the Shapley algorithm, derived from game theory. This algorithm allocates value to a touchpoint based on its ability to increase the probability of a conversion taking place, given it is included in the customer journey. This makes it fair. It is also data-driven, as it takes all previous customer journey data into account. The algorithm is based on A/B tests, making it interpretable. Economic efficiency also holds, as the results provide a complete overview of the (monetary) success of a campaign per media channel. All in all, a good attribution algorithm helps optimize marketing and media inputs by predicting outcomes of campaigns based on statistics and machine learning techniques.
How should it be implemented?
Configuring and maintaining an MTA algorithm requires a number of different resources. All relevant data must be collected periodically using Google Analytics or Adobe, for example. It should be placed on a server where it can be processed and analysed. The MTA algorithm will run on this server. The algorithm must be configured to work with the existing data structure, which is usually carried out by developers and data scientists. The data also needs to be checked for errors continuously and needs to be visualised in a dashboard (see our blog on data quality). Finally, it is important to have a dedicated team and a long-term focus within the organisation, which should be carefully considered before commencing any analyses. Isolated analyses often fail to lead to the collection of structural insights and changes in the decision-making process. For more information download these 10 Steps on Implementing Multi-Touch Attribution.
MTA is the only model that assigns value to touchpoints in the customer journey according to its contribution to the conversion. There is a standardised approach to MTA, with a classification system of what can be considered a good algorithm. Implementing MTA requires resources and dedication, but will provide you with invaluable insights and the ability to make data-driven decisions. Conversion attribution using an MTA algorithm is always preferable to basing decisions on last-click models.