algo learns wrong

When I worked at a reporting SaaS, I saw reporting data from hundreds of agencies. Thousands of accounts.

One pattern kept showing up. Campaigns that "stopped working" at week 3-4 with no obvious explanation.

At the time I thought it was normal. Audience fatigue. Creative decay. The standard explanations.

It wasn't.

Here's what's actually happening:

Week 1: You launch. Signal is mostly clean. The algo starts learning.

Week 2: A small tracking issue feeds partial data. Maybe a tag misfired. Maybe CAPI went stale. The algo adjusts quietly.

Week 3: Optimization has drifted 11-18% from reality. The algo is confidently wrong.

Week 4: You kill the campaign. Blame creative. Launch new assets. The new campaign inherits the same broken signal.

Cycle repeats.

I watched this pattern play out across hundreds of accounts before I understood what was actually happening. The campaigns weren't failing. The signal was drifting. And nobody was checking because the reporting dashboard didn't show it.

That's why I left a reporting SaaS to build Paid Media Lab. Because the gap between what platforms report and what actually happens is where most performance problems live.

If your campaigns keep dying at week 3-4 and you can't explain why, reply "drift" and I'll send you the 3 things to check.

Saku

PS: The GCLID chain on Google Ads is even more fragile than Meta CAPI. Breaks silently at form submit on most CRMs. Almost nobody catches it.