Vertical scaling killed my ROAS — horizontal saved it
I tried to scale a winner by cranking the daily budget. The algorithm punished me for it.
— The setup: Facebook, info-product, one ad set at $50/day doing a steady 2.3x ROAS.
— The move (wrong one): I 4x'd the budget to $200/day overnight on the same ad set, assuming linear scaling.
— The numbers: ROAS cratered to 1.3x within 48 hours. The budget shock reset the learning phase and the algorithm went hunting cheaper, worse users. That two-day stunt: $400 spend, $520 revenue, +30% but trending down fast and the ad set never recovered its old efficiency.
— The fix: I duplicated the winning ad set into six near-identical ad sets at $50/day each, different audiences. Horizontal. Five held 2.0x+, one flopped. Net at the new scale: $1,500/day spend, $3,150 revenue, +110% blended.
— The lesson: doubling budget on one ad set is a different action than doubling your spend across many. The algorithm treats a budget jump as a brand-new problem to re-solve.
What I'd do differently: Never touch a winner's budget by more than 20% a day; duplicate to scale instead. The $400 budget-shock experiment was avoidable. Numbers illustrative; the learning-phase reset is real.
Arb Files
@ArbFiles