The issue is many-fold: 1) formalising multiple half hour interviews into something you can A/B test is non-trivial, 2) given the logistics of these interviews the sample you end up with is small and therefore bias.
Having a user testing session is relatively rare, saved for special one-off products/features. Actually having feedback from those sessions distilled into product feature proposals, and getting them into the product roadmap is even rarer.
In the meantime, hundreds of A/B tests based on metrics like these would have been ran and resulted in a permanent change to the product. It’s just so much easier for everyone involved to trust “the data” vs something that requires nuanced interpretation.