Hello @julianno12,

Thank you for contacting us and sharing the detailed information regarding your concerns.

Geo-level model assumes a similar media impact mechanism across different geos which cannot be guaranteed for different brands, i.e., media spend of different brands may impact the target KPI differently. Though different geos impact target KPI differently, the hierarchical model gets the most benefit when hierarchies are more similar to each other, but they don't have to be identical nor does the model fail if geos are very different. Brands may have bigger differences than geos (especially since there's population scaling for geos), therefore, the basic assumption of a geo-level model that makes it better in estimating the relationships may not be satisfied when you consider brands as geos. Additionally, geo-level modeling would require you to define population for each geo, and the geos are assumed to be disjoint, i.e., an individual can only belong to one geo at a time which may not be the case when you use brand hierarchy. Keeping these core assumptions in mind, if you think the brand-level hierarchies are applicable for your use case, then you may experiment with the same.

Alternatively, creating separate models for each brand may be a good approach. To model the interrelationships of how all the brands’ marketing activities impact the target KPI of a particular brand (say brand_1) you may either add the media data of other brands as paid media channels (i.e., brand_1_tv_spend, brand_2_tv_spend, and so on will be separate columns) or as control variables. This decision should be based on whether you can control the budget allocation of brand_2 while modeling its impact on brand_1. If the budget allocations are fixed beforehand, i.e., you don’t intend to change the brand-wise allocation of your marketing budget, then it would be ideal to add the other brands’ marketing activities as control variables as they are considered non-intervenable.

Aggregating all the other brands’ marketing activities while modeling for a particular brand, say aggregating marketing activities for brands_2 to brand_n while modeling for brand_1’s sales, may be a good idea, especially if you have large number of brands. As you intend to use a national-level model, if the amount of data isn’t large enough to accurately estimate all the parameters in the model, you may choose to aggregate the rest of the brands which will reduce the number of parameters that need to be estimated. Aggregating would reduce model complexity and this would be beneficial with limited data and national-level granularity for the model. But this will make you lose the ability of attributing sales to brand-specific marketing activities which is a trade-off.

Do reach out if you have any further questions regarding this.

Google Meridian Support Team