Sure, you can do document classification with an LLM and it'll probably work for much broader documents, but running an LLM requires a machine with a lot of GPU compute power, while you can run a xgboost model on standard hardware, for the fraction of the time.