SQL vs NoSQL: how do you actually make the decision?

we've got this interesting problem at work where we handle both highly structured transactional data, like user accounts and order history, but also a lot of semi-structured stuff: activity logs, product catalogs with varying attributes, user preferences. for the structured part, postgres is a no-brainer. but for the semi-structured, we're debating between pushing more into postgres with jsonb columns, or bringing in a nosql store like dynamodb. the appeal of jsonb is obviously fewer moving parts, keeping data locality. but i'm a bit wary of potential query performance issues on complex json structures, especially with indexing. if we go dynamodb, we add operational overhead, but it might be better suited for schema evolution and certain access patterns. what factors do you all weigh when making this call? particularly when you're looking at a mixed workload and trying to avoid adding another database unless absolutely necessary.