Is there a way to get the partial pooling effect without fitting all the data?

Hello,

I have a timeseries sales dataset with about 19 million observations across five years for 658 different items.

For example, the item “Eye Trio” would be a part of the following hierarchy:
BEAUTY
COLOR COLLECTIONS COLLECTIONS-EYE
EYE COLOR
ARTISTRY
SIGNATURE COLOR

Is it possible to use any or each level of this hierarchy while only sampling a few items from each hierarchy? Could I sample the hierarchy using it’s actual mean and std to feed into the level below until it eventually gets to the item level. sales…which is what I want to model?