If your excellent analysis correctly infers that A causally increases B, this fact remains true without reference to the model and can be used in practice.

If your excellent prediction model finds that A predicts B, this is only true within the exact model you fitted.