Case #017: The $4 CPA that was secretly a $14 CPA

Case #017: The $4 CPA that was secretly a $14 CPA
The report glowed. Blended cost per acquisition: $4.10 across a sweepstakes offer, $1,200 spent, looked like a print machine. I scaled budget 3x in two days.

Then the money got worse the more I spent. By day 6 I was at $3,800 spent and the back-end revenue had barely moved past where it sat at $1,200. Something was eating the new traffic alive.

I broke the blend apart by placement. One placement — a single in-feed slot — had a $1.90 CPA and was carrying the entire account. Every other placement sat between $9 and $14. The blended $4 was an average hiding one hero and eight corpses. Scaling 3x didn't scale the hero. It poured budget into the corpses, because the algorithm chased cheap impressions, not cheap conversions.

The fix was unglamorous: I stopped trusting any account-level number. I rebuilt reporting so every placement, every creative, every hour-of-day had its own CPA column. Then I cut everything north of my breakeven CPA and rebuilt the campaign around the one slot that worked.

Final arc: paused the bloat, isolated the winner, scaled it alone. $2,600 spent, $3,950 back, 51% ROI by day 12.

— The lesson: a blended metric is a crime scene with the body hidden under the average — break it apart before you scale, not after.