And it's not like there wasn't any analytic work just before those 'feelings'. Some things are simply too different to even compare (how many city states can a player claim confidently for himself?). So at the end of the day, it's a work of fixing the biggest offenders (civs that win too often, victories that are not achieved frequently, culture trees that leads to victory too often, such things).
Probably, actual human games are not huge to make good statistics, but it's backed by your AI games, so everything is fine.