I would second exactly what @Judge said. Running through it with a CSM/SE is very valuable and running it for a test period will help you determine the proper threshold(s).
Aside from that, I would recommend one threshold for blocking, another range that is challenged and another range that is allowed through. The challenge section helps with the ML functions.
The only safe place to start, in my opinion, is to block scores less than 2…aka ‘1’. Outside of that, you’ll want to analyze traffic for a bit and customize.
I’ll assume you have an Ent plan with Bot Management so one other thing we did was log requests as various scores to help with analysis.
Ex. Rule 1: (cf.bot_management.score lt 2 and not cf.bot_management.verified_bot) action ‘Block’
Rule 2: (cf.bot_management.score eq 2 and not cf.bot_management.verified_bot) action ‘Log’
You could also place them in ranges, say 15-30, but I’d recommend being granular below 15, personally.