It's easier to figure out a way to scale something that is working than trying to fix a scalable approach that's falling flat.