The five that do work
Shots for and against, and above all their quality measured in xG. Clear chances created and conceded. Corners, which speak of sustained pressure. Fouls and cards, useful in their own markets. And minutes of possession in the final third, which separates the team that has the ball from the team that has it where it counts.
The four that do not
The winning run, which is already in the price. Head-to-head records from five years ago, with different squads. Total possession, which can be pure sterile passing. And the plain goals average, without distinguishing whether the team creates or is simply gifted goals.
The order matters
First the context: what each team is playing for and which squad it arrives with. Second the process: xG and chances over a window of ten or fifteen matches. Third the market: which way the price has moved. And last, the decision. Do it backwards and the statistics end up justifying the bet you already wanted to make.
Home and away do not get averaged
A team can create plenty at home and vanish on the road. Averaging the two conditions erases exactly the information that was useful. The same goes for surface, weather and long trips in continental competitions.
Sample size: the silent filter
Five matches tell you nothing. At ten, the process metrics start carrying signal. Below that, any conclusion is a story told with numbers. And in leagues that are just starting, the honest move is to wait rather than force a prediction.
How a model does it
A statistical model does exactly this but without tiring and without preferences: it looks at every variable, in every window, and compares against the price. Then it publishes its probability. That is the work the AI on the home page does, free and with a visible track record.