It's always a bad idea. Eight decades of software has shown it to be true, multiple books written on it. The inherent problem with it is the idea that writing a replacement is somehow going to be easy or fast. But if it were easy or fast, somebody would have done it already! Then come the additional problems of a rewrite like second system effect, inherent complexity, human error. It's impossible to predict how long it could take to finish making it, so fast-or-easy goes out the window.
Even though it's a bad idea, in very rare cases it can be the least-bad idea. But like you said, that takes an expert to figure out, and there's not a lot of experts floating around.
Even though it's a bad idea, in very rare cases it can be the least-bad idea. But like you said, that takes an expert to figure out, and there's not a lot of experts floating around.