Why? Were the statistics, etc, not useful?
I would assume that if the underlying models were already good, about the only thing an LLM might be good for would be filtering new info and/or presenting information in a useful way, both of which can be added to an existing flow?