Advanced Modeling Scenarios

Granular Models
Granular models are created for the purpose of generating granular MMM insights - at campaign or sub campaign levels. Example : Incrementality at specific TV Channels / Shows Level. Incrementality at various KOL Influencer level, etc...

• Granular models need to be run separately from the main model.
• Decomposed KPI data of the main model is passed into another model, along with the sub-component/factors of the independent variable to model for granular insights
• Purpose of having a granular models is to get granular insights only (i.e, when ifactor application is not possible for attribution calibration)
• We do not roll up granular models to master models (i.e, additive modeling is not supported for this)

Hierarchical Models

• A brand might want to run models across multiple geos / product levels
• Lifesight support hierarchical modeling across one dimension
• Our approach is to run separate models across the dimension and then aggregate the models to create one master model.
• Users can use the models separately for planning and optimisation for a geo/product, Or
• Users can use the national/brand level model (i.e, the master model) for overall optimisation

Nested Models

• Lifesight today run nested models based on one assumption - i.e BOF investment is not truly independent of TOF investments. (i.e, independence assumption of input variables are violated from regression in this context)
• Nested models are created to capture interaction effect between input variables
• If users want to capture any other interactions (other than TOF influencing BOF), they can let our marketing scientist know about this and we shall incorporate that into the model
• We will very soon make these assumptions of interactions transparent in the UI and we will let users update these "relationships" while building the model itself

Additive Models

• This is the option to merge multiple models into one
• In Lifesight this is not an automated process yet (as model merge needs a lot of data validation on the existing models)
• Example : We currently run separate models for Shopify , Offline , Marketplace revenue. If the user wants to make decisions at overall revenue level, we need to merge these separate models to one unified model.