Multiple channels of marketing are necessary for any firm to succeed but tracking down the individual contribution of a touchpoint is difficult. The multitouch attribution analysis helps firms keeping a track of the different channels. However, only about 32% of managers know about this powerful analysis method(iab.com).
What is attribution?
One of the most powerful changes in marketing has been the infusion of information technology. This enabled us to tap into different platforms of sales, promotion, retail, and customer service. This is done through managing the 4Ps of marketing mix or sometimes through marketing mix modeling.
However, when we spend our money on different channels we also need to audit their effectiveness. The process of linking each channel to a specific outcome is called attribution.
As an example let’s say that a person walks into a Samsung experience store. The person checks out the phone and decides to buy a new phone. However, the store does not keep the specific color that he desires. Consequently, he searches online on the Samsung website. He finds his model and color of choice available. However, before finalizing his purchase, he also checks out some reviews on social media. Consequently, he makes the purchase online through Samsung’s website.
Now, in this case, we will find it very difficult to attribute the sale to a particular channel. Was the store important in making his purchase decision? Or social media reviews? Attributing the conversion to a particular channel becomes exceedingly difficult when we have multiple touchpoints for the consumer.
Single touch Attribution versus multitouch Attribution
One of the simplest ways of understanding this particular journey is to simply assign the Attribution to a single touchpoint. One can say that either the visit to the store was extremely important and hence all of the credit should go to the store. Alternatively, one can also say that the website was critical and one should give the credit to the functional website.
We call this attrbution as single touch attribution.
Now let us consider the multitouch attribution analysis. In multitouch attribution, we assign a certain weightage to each of the different touchpoints. For simplicity, let us consider that we assign 40% weightage to the store and 40% weightage to the website. Consequently, we shall assign 20% weightage to the social media website.
This kind of assignment brings a bigger challenge. Firstly we need to have some kind of rule to assign specific weightage to specific touchpoints. These rules need to be static. Also, it is logically driven. Secondly, we also need to have a robust method of tracking users through their journey. The second challenge is taken care of by the different kinds of cookies and tracking ID.
Types of multitouch attribution
As we briefly discussed, there can be different ways in which we can assign weights to two different touchpoints. Now we will look at specific methods that can be used for assigning different multiple touchpoints.
- First interaction – This is a simple model. Here we give all the credit to the first touchpoint for the customer. It is fairly easy to understand but may not provide a true picture of the customer journey.
- Last interaction – Another model very similar to the first interaction. However, it is quite a common method as it is faster just like the first interaction.
- Second last interaction – As the name suggests, here we give all the credit to the second last interaction. This is commonly used for search engine clickstream data. Typically useful for inorganic search campaigns like Google Adwords.
- Linear attribution – all the touchpoints get equal weightage. It is also known as raw attribution.
- U-Shaped – In this attribution model, the first and last points get the maximum weightage of 40% while the rest 20% is shared between all other touchpoints.
- W-shaped or Full Path Model – In this attribution model, the first, third, and last stage are considered the most important. That is the visit, lead generation, and the final conversion stage. Typically, these touchpoints get about 90% of the weightage, while the other touchpoints get a total of 10%.
- Triangle shaped or time decay model – In this model, we assume that each marketing touchpoint has an impact for a limited time. Its effect on consumer decision diminishes. Therefore, we assign a higher value to the last touchpoint while assigning progressively lower value to the earlier stages. This assignment is done based on a mathematical equation derived from the hazard function.
- Custom model – These are the most advanced models of the lot. In this type of multi-touch attribution, we assign separate weightage to each touchpoint. This is the most trickiest and involving method. This is where all the advanced analytics and AI wizardry comes into play.
Methods for multitouch attribution analysis
- Bagged logistic regression method – This is one of the simplest methods for multitouch attribution analysis. Here we use a special type of regression model for assigning weights to the different touchpoints.
The basic premise of this model is that there are two kinds of customers. On one hand, we have those that end up buying the product, while on the other hand, we have those that do not end up buying the product. Therefore, we use a logistic regression model that is used for the prediction of this type of binary behavior. Secondly, we use the bagging technique. Bagging means that we use subsampling without replacement. You can read a conference research paper that discusses this method.
- Maximum likelihood regression – This type of model is also very similar to the bagged logistic regression model. However, the approach is different. Let us say, we observe a certain pattern from the data we have. This type of model attempts to provide the scenarios that would have most likely lead to the data we have.
For instance, let us that when we look at the conversion data and find out that there is YouTube influencer who gave rave reviews about our product. We notice that of all the conversions, there is a high proportion of people who have watched her video in their information search stage. On the other hand, we also notice that of all the people who were interested in the product and did not buy the product. Most of these people did not watch her review video. Intuitively we would give her a lot of credit for the success of our product. A maximum likelihood regression will also get to a similar conclusion. Albeit, with some higher reliability in the results than our intuitive model.
Link to the conference paper that discusses this model in detail.
- Survival function based attribution – This model is based on two concepts: survival function or hazard function and probability. Here, we assume that the most recent interaction will have a higher impact on consumer decisions than earlier ones. The memory, recall, and priming of touchpoint fade as time passes.
You may read more about this model in research by Wang and Dalessandro et al.
- Deep neural network model – As the name suggests, this model is based on the neural network. It is one of the most advanced models for a couple of reasons: Firstly, some of the neural net-based multi-channel attribution models are more difficult to develop. Secondly, the results are easy to interpret, however, there can be issues with transparency of the model as well as tractability. This type of multitouch attribution model can be quite helpful if we want to develop a custom model for multitouch attribution analysis.
- Game theory model – This is also a unique model that can be used for MTA. Here, we may use a cooperative game developed by Lloyd S Shapley. In this model, the contribution of different touchpoints is distributed as a game theory output.
You may refer to this post on Medium by Jacky Yuan to learn more about this method.