Hyperparameter tuning has always proven to be effective for increasing model accuracy - however this can be expanded upon. When trying to predict on time-series data, sometimes the dates in question are the most effective indicator of volume - especially in the topic of trends in customer behaviour.
extended-prophet aims to combine the effects of hyperparameter tuning with a selection of exogenous regressors - picking the most optimal patterns in dates to use as a basis to futher increase model accuracy.