What happens if the analyst wants to analyze the data before it gets into Redshift? Also, what do you do when the underlying data's schema changes (i.e., the API tier gets upgraded and new fields appear in S3 but not in Redshift).
For QA:
The MySQL ETLs are brittle. They depend on a certain schema. If it changes, part of the ETL will break.
Sometimes definitions in the logs change. So far everytime this has happened and I wasn't notified, their effects are quite clear. Ex a metric doubling on a certain day.