YepIMO...
There's a problem with analytics in sports, in that it only considers the raw statistics with vast sample sizes, that bear out over the course of an entire season.
They do not factor in the individual variables of a skill-based game composed of athletes with various levels of energy, mental acuity, or injury status. They can't because that shit would have to be calculated on the fly, before every play. We'd need some serious AI for that (wonder how far away we are from the first AI GM/HC...but I digress).
So the NFL may be 50/50 on 2-point conversions, but the right side of your O-Line may be struggling that day against the D due to your starters being out. Suddenly that 50/50 chance that bears out of over the course of a season isn't so 50/50. Conversely, your offense might have kept their Defense on the field all game, and they are tired, and suddenly you might have a greater chance than just 50/50. And those are just current game examples, not including season long factors like who has the best offense or defense to begin with.
The old "situational modifiers" are not considered from a pure analytical standpoint, and it's always been that way, which is why right vs lefty, lefty vs right shit doesn't always work in MLB during the playoffs either.
Playing the odds is normally wise, but a good coach/manager has to consider the current flow of the game, otherwise their analytics will blow up in their faces at the worst time because they rolled a 1 on their d20 when they only needed a 5 to score a hit.
So some geek saying "go for 2 down 14" because it bears out that there is a 12.5% advantage is right in a general sense as for best practice over the course of the season. But saying to do it "always" on any given gameday, that might actually be the worst thing to do.
Shitload of other variables, hundreds at least
A successful extra point does not lose momentum after a TD
Failing on a 2 point conversion does lose momentum after a TD
That is only one variable that I am sure isn't factored into the statistics