The payoff from having some engineers doing the T of ETL (transforming the data to be more queryable, denormalizing values etc) is insane if you have problems in democratizing data analysis/reporting

The dumb thing for people who don't know what this looks like: make a separate database. Create tables based on what you often want to query (but don't try to keep the same shape as the base data!). Then figure out how to fill in those tables from the base data