What metrics would you track to ensure a newly deployed AI summarization feature is successful at nference?
