There exist metrics like ELO or GLIKO that can be used to rate the overall performance of teams. This is usually a good predictor for the overall winrate of a team A. But doesn't predict as well whether A will win against a particular B. This is because there is a lot more going on in particular matchups that is difficult to capture mathematically. It is not uncommon to see A beating B consistently, B beating C consistently, and C beating A consistently.

Perhaps a neural network can be trained on a lot more data to make predictions for each match, and I imagine who gamble a lot have probably done things like this already. But then you don't have a number to optimize.