<b>Case #002: The day scaling killed a winner</b>

<b>Case #002: The day scaling killed a winner</b>

We had a clean winner. Nutra, a stubborn skincare angle, sitting at 140% ROI for eleven days straight on $90/day. The temptation when something works is to feed it. So we fed it.

Day 12 we went from $90 to $400 in one jump. Classic mistake, and we knew it, but the numbers were so good we gambled the learning phase would survive.

It didn't. CPA leapt from $14 to $31. The algorithm, handed 4x the budget overnight, re-entered learning and started buying the cheapest, worst traffic it could find to spend the money. We torched $1,200 over two days chasing the version of the campaign that had already died the moment we changed it.

— Pre-scale: $90/day, $14 CPA, 140% ROI
— Post-jump: $400/day, $31 CPA, -8% ROI
— Recovery cost: $1,200 burned, 3 days lost

The fix was boring. We rebuilt the original ad set, left it untouched at $90, and duplicated it into a fresh campaign at $120 — a 33% step, not a 4x leap. Then we waited 72 hours before the next 20% bump.

By day 20 we were at $310/day and back to 118% ROI. Slower than the gamble would have been if the gamble had worked. But the gamble never works.

The lesson: budget is a learning signal, not a faucet — 20% steps every three days beats one heroic 4x every time.