In my current project, I started using a SMT solver to type a language I'm compiling (find out the "best" types for every expression). Then I also used it for storage ressource allocations (finding out which programs should store their output so that later retrieval is the cheapest), and then again to layout some graph on the GUI; and I have another similar use case already in mind.

Of the many ways to solve a complex problems that we have on our tool box -- algorithms, neural networks, constraint solvers -- I would totally agree to say that solvers are underused. But when you get used to them it's not hard to find use cases.