In his talk about "milestones in the development of lazy, strongly typed, polymorphic, higher order, purely functional languages" David Turner mentions that the wan't adverse to SASL (https://youtu.be/QVwm9jlBTik?t=1819) and Miranda (https://youtu.be/QVwm9jlBTik?t=2330) —both predecessors of Haskell—to be used in industry. Getting out of the ivory tower was not a last minute idea, it seems.
The Peter Landin paper "On the Mechanical Evaluation of Expressions" (https://www.cs.cmu.edu/~crary/819-f09/Landin64.pdf) makes it totally clear that, for Landin, FP was an alternative to the bookkeeping required in systems programming.
Since he stands at the beginning of the tradition that led, via Turner's work, to Haskell, I think it's not a huge leap of the imagination to attribute Haskell's lack of success, as a language for building systems, to this inherited attitude that the programming system should be elegant, principled and mathematically structured. No concessions are made to the practical needs of someone writing, for example, an operating system, except to the extent that they provide an occasion for a new theoretical construct. (Lenses are a recent example of this.)
And take implicit data structures, for example. I know Edward Kmett has explored this idea in Haskell, but really, it's a totally alien concept for the FP philosophy. Just as traditional operating systems tend to require a little bit of assembly code in addition to C, the purist FP system that wants to manipulate genuine implicit data structures will need to call on some outside language with the power to manipulate them... That seems like an intentional state of affairs.
