[dllthomas]

Haskell can actually make a pretty good imperative language (and you can put together little imperative sub-languages as well). That aside...

There's a reasonable perspective that says the type of problem that's best solved in a language is a problem for which there's good library support. For that, I'd probably recommend scanning https://github.com/Gabriel439/post-rfc/blob/master/sotu.md

But in my experience, where Haskell really shines is sketching out operations on some type that I know I'm going to get wrong - substantially or subtly - my first many tries. When I go to fix it, the compiler helps me find everything I need to change in tandem, and feel ahead to find inconsistencies in my model before I get there in the code. One example of this kind of problem is compilers - where we parse into an AST, transform the AST, and then produce other things from the AST. As development goes on, the AST evolves, and you have a lot of help knowing what needs to change to match and what you can ignore. I've recently been doing this for SQL (though with the goal of analyzing the queries, not producing an executable).