This. Determining when the effort should be applied is the tricky part. Games are still hit driven and get cancelled/re-purposed during development. You can spend a lot of QA engineering time developing systems to test functionality that never ships (case in point would be Fortnight - the original shipped game did not need to be tested against the current 100 player game instances and huge load but they could have spent a bunch of time testing AI systems that are no longer any part of the game).