The only reason to ever use a non-relational db is for scalability and performance reasons. Joins and transactions are hard to do correctly and efficiently in a distributed system. So “NoSQL” solutions can be a good fit if your data is too big to fit on a single host and you can get by without joins and transactions.
(This is a massive oversimplification, but still used rule of thumb.)
And most companies vastly overestimate their data, and believe it to be "big", when it could be trivially handled decades ago by server-grade hardware.
> The only reason to ever use a non-relational db is for scalability and performance reasons.
And, most importantly, lack of market availability. Nobody is going to sell you a relational database nowadays, and rolling your own is... daring. Postgres was the last serious kick at a relational database, but even it eventually gave into becoming tablational when it abandoned QUEL and adopted SQL.
But tablational databases are arguably better for most use-cases, not to mention easier to understand for those who don't come from a math background. There is good reason tablational databases are the default choice.
(This is a massive oversimplification, but still used rule of thumb.)