My thought experiment is this: take a simple, single page app that take user information, runs some calculation, and returns the results. Add login/SSO/or what ever third-party integration. It will be faster to bring value to your customers using a dynamic language (Ruby/Python/PhP), then using (Haskell/Scala). I don't think you really see the benefits of Hindley-Miller type inference, unless you are building complex services (think: compilers, real-time allocation services) that require multiple layers of domain specific abstraction. Advantages of using strongly typed languages for simple SaaS often doesn't hold, and there are very few, if any? successful SaaS start ups that have taken this route. I just think not moving fast enough will kill your start up before you ever get to the point of having to worry about 'tech debt'

Note: This could also be true for programming tooling/environmental reasons, and strongly typed languages do help smaller teams handle larger codebases/reduce tests!