I think it’s very hard to assess it quantitatively. Sema (semasoftware.com/ai-cod…) is doing good work on that but this whole space is still early

Qualitative data is still probably what matters the most