Drop in Media Attribution After Adding Seasonality and Trend to Model

[shantisb]

I’m noticing a significant drop in media attribution after adding Prophet’s seasonality and trend components to my model. The model structure and media inputs remain the same, and the overall model fit is still good in both cases—but the media contribution drops considerably when seasonality and trend are included.

From a business perspective, this level of attribution doesn’t align at all with what's observed operationally. We know that media plays a substantial role in driving performance.

Would appreciate your thoughts on:

1. Why the inclusion of seasonality and trend might cause such a substantial and erroneous drop in media attribution?
2. Any best practices for maintaining business-aligned attribution while accounting for time-based patterns

Thanks so much :)

[AdimDrewnik]

Probably because Prophet trend and seasonality are extracted from the raw data and not as a part of the model.