This recalls me something. Someone was fighting to reduce a source of uncertainties on a prediction from 10% to 5%. It is nice, in principle, except when the dominant source of uncertainties corresponds to a 50% error.