The cost of refactoring

Today a colleague was proposing a design change to the backend of a GUI heavy application with the main rationale that the backend would be more elegant and the change would be easy to implement. I told him that the real cost of refactoring is the necessity to repeat all the manual UI tests (our automated Macro Scheduler tests are not ready yet). A GUI application has many paths to failure and we have been developing this particular application for a couple of years now. Changing the backend would certainly break some edge cases and from the users perspective, the app would have regressed.

The only acceptable reason for a backend refactoring would be if we had lots of functionality to add in the short term and the change would make such additions a breeze, being a net saving in effort, in addition to being elegant.