They actually don't, they all (KenPom, NET, RPI/BPI, etc.), have average predictive value year to year (use them to fill out your bracket and you'll likely finish in the middle of the pack).
At this point, it is because the 1) the inequity of schedule, 2) few games/little data, 3) coaches experimenting with rotations, and 4) growth/injury (teams are improving quickly or not).
This is why a team can move 7 (or more) positions after just one game.
After conference play it comes down to a teams mean ability and their variance around that (the higher the mean and less the variance, the better), coupled against a one-and-done conference tournament or NCAA tournament platform.