My conversion pixel fired 90 seconds late and the algorithm went blind
The campaign was learning from the wrong signal the entire time. I didn't notice for two weeks.
The setup
— Conversion-optimized social campaign, $22 payout, ~$1,300/day.
— Postback chain: network to my tracker to the ad platform. Three hops.
The move
ROI was mediocre (~8%) and weirdly unstable. I traced a single conversion end to end with timestamps.
The numbers (illustrative)
— Postback took 60-120s per hop; total delay often 3-4 minutes.
— ~25% of conversions arrived after the platform's real-time attribution snapshot, so it optimized on partial data.
— After fixing the chain to sub-5s: ROI climbed to 29% in a week, same budget.
The lesson
Algorithmic optimization is only as good as the speed and completeness of your conversion signal. A slow postback means the machine is learning from a blurry, half-missing picture and confidently optimizing toward the wrong people.
What I'd do differently
Latency-test the full postback chain before scaling, not after. The plumbing is invisible until it quietly caps your ROI, and you'll blame the creative for weeks.
Arb Files
@ArbFiles