SIS’s win probability model is primarily a neural network trained on the last four years of NFL games to predict the results of a game from any game state.

For a given matchup, it uses a measure of opponent-adjusted team strength that relies on recent Expected Points Added performance.

Each team’s last eight weeks both offensively and defensively are evaluated against their opponents’ performance in other games to get a more stable measure of the recent quality of each team.

The difference between these overall team ratings is used to estimate a matchup-specific win probability.

This is different than the full SIS pre-game expected score model, which is not available publicly and incorporates more specific factors, but this is a good representation of how we expect these teams to perform when matched up.

For the AFC Championship Game, we have the Chiefs with a 71.7% chance to beat the Bengals.

For the NFC, we have the Rams with a 54.9% chance to beat the 49ers.