Here’s how it works:
For each player, our technology matches them with a number of historical players, all of which share important characteristics with the current player. In the case of a QB, we might look at the ratio of TDs to INTs, or the ratio of long passes thrown to short passes. This allows to find historically comparable players, and assign a numerical value to that similarity.
For each game, we look at each team playing in the game, and again look historically at what teams have run similar styles of offense, defense, game strategy, and so on. This allows to find situations where similar offenses have gone up against similar defenses to the game we’re trying to predict.
Once we have a good grasp of these similarities, we now look at the combination of the two – similar players playing in similar situations. This also has a similarity score, one that is composed of how similar the player and the matchup both are.
Now that we have all this data, we run it through our projection algorithm. This generates a super-specific, data-driven projection based on game after game after game of similar players and similar defenses, all weighted by how similar they are to the present. You can use these projections to decide who to start, who to drop, who to trade for, and so on. In short, they help you win.