It’s true that football is a more of a random sport than other sports due to its low scoring nature, but from a predictive modeling standpoint I would say football analytics has really reached the potential of the tools available.
At this point it seems like the biggest gains are still at the feature/data engineering level, because a lot of potentially revealing metrics have not been been defined or standardized. Caley’s description of creating features for tracking fast breaks exemplifies this. When a lot of domain or technical concepts in soccer like ‘pressing’, zone coverage, or ball-carrying are tracked by metrics designed to track such concepts, we will start to be able to align the pedagogical views of football theory with empirical observations of them from a modeling perspective.
Re using ‘big data’ or tracking data, I would look to see whether IoT is making meaningful discoveries because they seem much more invested in analyzing and using that information than sports.