The loser funded the winner. The trick was treating dead spend as paid data, not a write-off.
The setup
— Broad push test across 6 GEOs, 5 angles, mobile + desktop, one big ugly campaign.
— Two weeks, $2,100 spent, $0 profit. A clean failure.
The move
Instead of deleting it, I treated the corpse as a dataset. Sliced every dimension, looking for the one cell that was secretly positive.
The numbers (illustrative)
— 1 of 60 segments (Germany + angle 3 + Android) ran at +70% ROI, drowned out by the rest.
— Rebuilt a tight campaign around just that cell: scaled to ~$700/day at 55% ROI.
The lesson
A failed broad test isn't a loss, it's a map. The $2,100 bought me coordinates I could never have guessed. People who win at scale aren't luckier, they're better at mining their own failures.
What I'd do differently
Budget the exploration phase as a line item, not an accident. "I will spend $2k to find one winning segment" is a strategy. "I lost $2k" is the same money with worse morale.
Arb Files
@ArbFiles
My most profitable campaign started as a $2,100 loss I refused to delete
This post was published in the Telegram channel Arb Files (@ArbFiles). All its posts are on the channel page.
Open on Telegram