Sounds like a great use case for an automated ML cleanup/reporting feature. Maybe as a daemon as a bolt-on fix, or integrated as a feature into the support software itself.