A/B testing auth methods is tricky. It's fine for "can we get better sign up rates" but it wreaks havoc on "can my previous users still sign in". At best you can AU/NA test - show different pre-auth treatments in different, very distant geographies that nevertheless have roughly similar user characteristics.

I don't disagree with you and it was proposed. However we didn't have A/B infrastructure in place then and on a project that may never make money it just wasn't a high enough priority to justify the spend.