The setup. E-commerce dropship via paid social, $34 AOV product. Illustrative numbers, painfully real habit.
The move. Every time a campaign dipped I'd edit the budget, swap an audience, or pause-and-relaunch. Felt active. Felt like work.
The numbers. Across a month: 31 significant edits, each one tossing the optimizer back into the learning phase. Learning-phase CPA averaged $19; stabilized CPA was $11. I spent an estimated $4,200 of the month's $16,000 budget perpetually in the expensive learning window — money I lit on fire by fidgeting. Blended ROI 0.96. A control campaign I forced myself not to touch: 1.31.
The lesson. On algorithmic sources, intervention has a cost the dashboard doesn't label. Every edit resets the model's confidence and re-buys expensive exploration traffic. Discipline beats activity.
What I'd do differently. Set a rule: no structural edits inside a 72h window unless spend has 3x'd the target CPA with zero conversions. Boredom is not a reason to touch a campaign. I keep a tally of edits now — if I'm over 2/week per campaign, I'm the problem.
Arb Files
@ArbFiles
I reset the learning phase 30 times and called it optimization
This post was published in the Telegram channel Arb Files (@ArbFiles). All its posts are on the channel page.
Open on Telegram