xG: the model that taught computers to value a shot
xG (expected goals) gives each shot a number between 0 and 1: the probability that it ends in a goal. A shot from very close to the goal gets a high number; one from far away, a number close to 0.
That number does not come from intuition. It is calculated by a statistical model, often using machine learning techniques, trained on a large number of past shots. The model learns which features make a goal more likely: distance, angle, body part, the type of previous pass or whether it was a set piece.
Adding up the xG of all of a team's shots gives an idea of how much it “deserved” to score. If a team scores far less than its xG suggests, it may have been unlucky, or have a wasteful striker.
But it is worth knowing that each company has its own model and the values do not match. And an xG without data on the defenders' positions does not see the whole context. It is a tool to understand better, not an absolute truth.