When do we need to prove (with prototyping or any other kind of proof) that a design solution that we came up with is good before implementation?
Does it depend on how much the actual implementation would take in comparison to the proof of concept? Does it depend on whether there's any uncertainty about the effectiveness of the solution?
Example:
A process that currently runs in one server can possibly be optimized with parallelization across multiple servers. It is not clear whether the performance gain would be significant - not everything can be done in parallel, some stages of the process must still be done serially. If the overall performance gain is not significant, it would not worth implementing the change.