This could slow down queues slightly but would save a separate dedupe process - maybe we can quantify
We are doing this now - albeit on a basic 'quick' method