TL,DR: Polls aren't trying to predict future performance, they're assessing the resume of teams. That's a completely different goal than what Vegas/analytics have.
Two easy examples from this season.
BYU beat Baylor in OT at home. From a resume-based polls perspective, that's a big BYU win. From a forward-looking analytics/Vegas perspective, that's a Baylor win.
Same story with Utah/Florida. Utah can't be that high in the polls because they lost that game. But analytics/Vegas will consider Utah coming up yards short on the road to indicate Utah was the stronger team. This explains why Utah is so high in algorithmic rankings this year (same for BYU in 2020 when we barely lost to a good CCU at their field on short notice).
If you're trying to predict future games, then Utah losing away to Florida as time expired is probably at least as strong as Georgia struggling with Mizzou. But from a resume perspective, which focuses on binary results, it's obviously much worse.
And there are 3 ND or BYU games off the top of my head where the resume and the forward-looking mindset diverge. BYU/Baylor as mentioned, BYU/USU where we played on par with UNLV, and ND/tOSU where ND kept it reasonably close to one of the super-elites