Using RDS to read/write/store data is slow and not portable. We may want to switch the data pipeline or dashboard to being written in Python in the future, so using a format that Python can easily read would be preferable.

Looking at a comparison of different formats and packages, storing as a CSV and reading/writing using data.table via fread and fwrite seems pretty good. feather could also be an option (is supported in both R and Python) but it's not clear if standard dataframe and dplyr procedures work with it. data.table data can be seamlessly processed either with dplyr or with faster data.table syntax.