As the piece above points out, people directly involved in football can perceive the effects of mood and affect on performances in a playing group.
To a data analyst, a striker looks like he scores 10% of his chances. To a coach, he looks like a guy who finishes 20% of his chances, but his dad died last season, he got divorced in the summer, and he had to recover from an injury after November. To xG as a measure, he looks like "a striker" standing in an area on the pitch and his particular finishing ability often isn't even considered.
These measures such as xG used so blithely in predictive models and simulations are just not that great. If you can mount a critique of using goals themselves as your basis for prediction, you can mount a very similar critique of xG, even while agreeing it improves on goals.
