Automated filters will just drive determined botters to play the system and perfect their craft until they can no longer be automatically identified, in my opinion. I’m more of the stance that accounts should be reviewed manually so that a leap into convincing bot accounts will need to be much more dramatic, and therefore difficult. If it’s done the hard way from the start with staff who know how to identify these accounts, it may keep it from growing into an issue to begin with.
Any threshold to be automatically flagged for review should be relatively low, but the process should also be quick and efficient. Adding more metrics to the flagging process only means botters will have a narrower gaze to avoid. Once they start crunching the numbers and streamline mimicking real user accounts it’s game over.