We did an analytics engine to pick QBs as draft picks.
It was tremendously good at figuring out if someone would be a bust vs a success if you put past draft picks in it blind. The model vaulted guys like Joe Montana to the top and people like Akili Smith to the bottom, if you put them all in the same pool and did not disclose they were past players.
A GM of a sports team did use it, but not for football.
What I can say is that the #1 predictor of success for us involved comebacks in the 4th quarter against higher-ranked teams or ranked teams (different points for each, higher-ranked teams were more point value). Of course ranking is subjective, but even so - it was tremendously accurate.
(It was a factor of opportunities vs success rate, but scored so that larger sample sizes looked better. QBs from dominant teams then didn't always rate as high, so you missed out on guys like Tua but the upside is that you missed out on guys like Tua!)
Now, did it work when applied to QBs moving forward and not just blind historical data? I have no idea.
But I expect the comebacks factor is still a great indicator.
If you want to judge a QB's ability to contribute in the NFL, start with that.