Most of the selling points on scala-lang.org still apply over C# - type inference, pattern matching, case classes (C# makes these less bad than Java, but the field/property distinction can still bite you), an inheritance model that supports the good parts of multiple inheritance without the problems, covariance/contravariance, for/yield syntax that's generic and user-controlled, higher-kinded types.
It takes a while for your code to get big enough for the last part to matter, but it's huge when it does - you can handle any kind of "context" in a generic way, e.g. async calls, error handling, audit logging, software transactional memory, or custom things for e.g. database access. F# has a number of these specific implementations but they're kind of "hard-coded" in the language rather than fully extensible. So I guess the pitch is: imagine C# async/await could be "just a library", and other contexts that you wanted to handle similarly could also be a library, and the resulting functions are first-class citizens in the type system (no awkward choices of whether a particular function is async or not, you can handle that generically).
It takes a while for your code to get big enough for the last part to matter, but it's huge when it does - you can handle any kind of "context" in a generic way, e.g. async calls, error handling, audit logging, software transactional memory, or custom things for e.g. database access. F# has a number of these specific implementations but they're kind of "hard-coded" in the language rather than fully extensible. So I guess the pitch is: imagine C# async/await could be "just a library", and other contexts that you wanted to handle similarly could also be a library, and the resulting functions are first-class citizens in the type system (no awkward choices of whether a particular function is async or not, you can handle that generically).