It is a lot easier for one unified ML model to remember all the conversations as opposed to a team of engineers who may not communicate with each other anywhere nearing perfect accuracy.