In this talk, I focus on a key challenge of designing a Multi-touch attribution solution - Identifying which methodology is best for this problem. This is a thorny issue as the variable we are trying to study isn't directly observable. I’ll discuss a novel approach to tackle this problem through simulated marketing environment.
Multi-touch attribution is a popular marketing science challenge. This is a tricky problem to solve as the goal label we need to train for is a latent variable i.e. can’t be directly observed in the real world. The e-commerce industry is actively looking at using fancy ML techniques to design a solution.
However, the question still remains, how do we know which methodology is the best for Multi-touch Attribution?
At Skyscanner, we have explored this problem space further. We use a simulated marketing environment to benchmark different methodologies. In this talk I will discuss:
An overview of this talk can be seen in this blog and this research poster