Path configurations are used to customize the way each user’s path is constructed from individual sessions. By customizing how paths are constructed, ChannelMix allows you to target specific business questions with your attribution model - which is trained on these reconstructed paths. ChannelMix supports two types of configurations: path compression and the direct filter.
When path compression is enabled, sequences of identical interactions are compressed into a single interaction. The compressed interaction is set to have the same timestamp as the first of the compressed interactions. If a path has two compressible sequences of the same channel, they are compressed independently to leave two interactions occurring at the appropriate places in the paths.
This is best explained through an example. Imagine you have a user that comes to your website four times: Paid Social > Paid Social > Display > Paid Social. Path compression would remove the duplicates to create a path with three interactions: Paid Social > Display > Paid Social.
By compressing paths, you allow the attribution model to focus on the interaction between channels. This gives you a much better understanding of how your marketing channels are working together to create target events. Compression also works to reduce the effects of bot traffic to your site - which usually appears as long sequences of identical interactions occurring in a very short period of time.
ChannelMix allows you to view your attribution models with path compression turned off as well. In this case, the model is able to learn about channels that users continually engage with along their path to a target. This provides a more holistic view of all customer paths, but does put you at risk of influence from bots and the rogue user that just loves clicking on the same ad repeatedly.
The direct filter adjusts the way customer paths are reconstructed to exclude direct interactions from paths. The only exception to this rule is for paths containing only direct interactions. These paths are reconstructed as a path with a single direct interaction.
As an example, you may have a sequence of interactions like: Display > Direct > Paid Social > Direct > Direct > Target. The direct filter would reconstruct these interactions as the path: Display > Paid Social > Target.
The direct filter is used to shift credit away from direct interactions and towards your marketing interventions. These are the interactions you can directly effect making them the interactions of most interest.