Scale once the adset has produced around ten genuinely profitable conversions. Ten is where there is enough evidence to act, fifteen is stronger, and twenty is high confidence. The number is not universal, it tracks how scarce and how valuable the conversion event is.
For a booked-call funnel, five calls is enough to start, because each call is worth far more and arrives far less often than an ecommerce purchase. For a low-ticket product, ten purchases is a thin sample and waiting for fifteen or twenty costs you little. The principle is the same in both cases, you want enough events that a single additional conversion cannot swing the picture.
Profitability is the gate, not volume. An adset with twenty purchases sitting at break-even is not ready to scale, because pushing budget into it makes facebook enter more competitive auctions and performance usually softens rather than improves. Scale things that are comfortably above target, so they have room to give some of it back and still clear the bar.
Do not use facebook's learning phase label as the trigger. Exiting learning is not a signal that anything got better, and there is no observed jump in performance at fifty conversions. Use your own profitable conversion count instead.