Predicting monthly vs predicting daily and aggregating monthly

Hi,

Thanks for the wonderful in prophet! It is very convinent to use.

Currently I have a dataset that consists of daily sales data, it is recorded manually so there are not much records during the weekend. My goal is to predict the sales monthly, so I wonder what is the difference between the two approachs:
1. Suming the sales monthly and predict monthly
2. Predict daily and aggregate the prediction monthly

I have tried the two approachs. For daily prediction, the smape metric is terrible, but summing them and validate monthly the result is acceptable. For monthly prediction, the result is ok but a bit worse than suming monthly.

Would you mind to explain a little further for this?