I just started incrementally throwing views at our existing schema, mostly lifting SQL out of the code base and committing it to the dbt repository. From there,

I just started incrementally throwing views at our existing schema, mostly lifting SQL out of the code base and committing it to the dbt repository. From there, I started heavily annotating the columns, and starting to write unit tests for the views. This uncovered a ton of hidden assumptions and dirty data which I either cleaned up, or manually flagged in the views (as in, skip the orders listed in this CSV file because they stem from the period of time where shipstation blablabla). Since then I'm continuously adding to it, refactoring the macros as if they were code, and it really feels in line with the normal workflow of a developer.