They do it because the A/B test results also matched their revenue improvement results.

I don't think many people involved actually believe this or actively think this way.

That being said, the result does have some similarity to how the physical world operates in large gatherings like primary schools or military, where people of similar attractiveness and few differences in their supportive ability would congregate. People figure out where they are on the consensus-attractiveness scale and match with people in the same range. This doesn't match how the world outside of that works, and therefore almost all guys and unattractive women have the same result on dating apps. For dating apps they haven't figured out a better solution, and don't want to or need to as it is also a perfect frustration feedback loop for them.