Scale gradually rather than in large jumps — commonly in increments that let the algorithm re-learn without fully resetting, spaced a few days apart rather than doubling budget overnight. Large, sudden budget increases often push a campaign back into a learning-like state and temporarily spike CPA right when you most want stability.
Horizontal scaling — duplicating into new audiences, placements, or geographies — often preserves performance better than just pouring more budget into the same ad set, since it avoids saturating the same audience. Watch frequency and CPL closely during any scale-up; both will tell you if you've pushed too far too fast.