According to a presentation I heard two month ago it's not good enough yet. They can make accurate predictions in many cases, yet not in all/enough. At the moment they see more potential in analyzing health insurance claims, as falling back to a human expert is easier.