I killed a winning ad on day 2 — it would have been my best

I killed a winning ad on day 2 — it would have been my best

The most expensive mistake in this batch isn't a campaign I ran. It's a winner I killed too early on noise.

— The setup: 4-creative split test, $25/day each, $100/day total, I made kill decisions after 24 hours.
— The move (the error): creative C had 1 conversion on 60 clicks at hour 24 and looked dead, so I paused it. Creative A had 3 conversions on 50 clicks and looked like the winner.
— The numbers: I scaled A. By day five A regressed hard to 0.9x ROAS — its early run was small-sample luck. I relaunched C out of desperation and it settled at 2.6x over a real sample of 800 clicks. The day-2 'winner' A lost me roughly $400 in scaling; the 'loser' C I'd killed was the actual money-maker. One conversion on 60 clicks tells you nothing.

— The lesson: at 50-60 clicks you have noise, not data. Killing on tiny samples means you're flipping coins and scaling whatever landed heads twice in a row.

What I'd do differently: Set a minimum sample — at least ~100 clicks or 5+ conversions per variant — before any kill or scale decision. My itchy trigger finger cost the best creative a week and ~$400 in wrong scaling. Figures illustrative; the kill-on-noise trap is the quietest profit-killer in media buying.