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

our team is trying to decide on the right database for a new core service. we have a mix of data: highly structured relational data like user accounts and transactions, but also a lot of semi-structured activity logs, feature flags, and product catalogs that change frequently. we're currently using postgresql for most things and mongodb for some flexible schema data. the question is, can we consolidate? is it better to use postgresql with jsonb columns for the semi-structured data, or should we stick with a dedicated nosql store like dynamodb for those parts? what are the real-world trade-offs in terms of query performance, development velocity, and operational complexity when choosing between these two approaches for mixed workloads?