Case #035: The winner that expired on a calendar

Case #035: The winner that expired on a calendar

This is a post-mortem on a campaign that did nothing wrong and lost money anyway.

A gifting offer — flowers and personalized goods — had run beautifully through early February. $8,000 spent, $14,500 back, 81% ROI in the two weeks before Valentine's Day. I scaled hard into the 13th and 14th, confident the trend would hold.

It held until 11:59pm on February 14th, and then it fell off a cliff. February 15th: $1,900 spent, $640 back, a 66% loss in a single day. The 16th was worse. I'd scaled a seasonal campaign into the exact moment its season ended.

The brutal part: the algorithm had no idea the holiday was over. It kept spending against learned demand that no longer existed. By the time I cut it, the post-holiday days had given back a chunk of the gain:

— peak (pre-14th): +81% ROI
— post-holiday tail (15th-17th): $4,800 spent, $2,100 back, 56% loss
— net for the full arc: still positive at +38%, but I'd left a clean 80% campaign and dragged it into the red

I made money and still managed to lose, by refusing to read a calendar.

The lesson: a seasonal campaign has a hard expiry date the algorithm will never see — schedule the kill before the holiday, not after the loss.