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

i often see heated debates about SQL vs NoSQL, but when it comes to actually making the decision for a new service or feature, it feels more nuanced than a simple either/or. we have both highly structured data like user accounts and financial transactions, and then more semi-structured or rapidly evolving data like activity logs, product catalogs, and user preferences. for the latter, i'm torn between using a traditional relational database like PostgreSQL with its JSONB capabilities, or going with a dedicated NoSQL store like DynamoDB or MongoDB. when do you typically decide that JSONB in PostgreSQL is sufficient, and when does it make more sense to introduce a separate NoSQL store? is it primarily about query patterns, scaling needs, or developer familiarity? often, a hybrid approach seems necessary, but how do you decide where to draw that line?