The name of the game is validation. Spend the time to build out a robust system for identifying risk and determining cost. To do this I have a series of test automation schemes and an architecture model that allows for rapid large scale refactoring. I find if I want to test some ambitious new idea the best way is to first prime the environment by consolidating existing functionality that scales more efficiently. It’s one of those counter-intuitive things where spending lots of time on refactoring yields fruit only in the second and third order consequences but the savings is on the scale of magnitudes.