The scariest tracking failures are the silent ones.
Nothing crashes. No error emails. The ads keep spending, the dashboard keeps showing numbers — and the numbers are quietly wrong. On one account I manage, a Meta Pixel broke in exactly this way. For weeks, attribution slowly bled out while everything looked normal.
Here’s how the diagnosis actually went, step by step — because the process matters more than this one story.
Step 1: Trust the smell, not the dashboard
The first signal wasn’t an alert. It was a mismatch: the ROAS in Ads Manager didn’t line up with actual orders in the backend. Small gaps are normal — attribution windows, returns, timing. But this gap kept widening.
Rule one of performance marketing: when two sources of truth disagree and keep disagreeing more, stop optimising and start auditing.
Step 2: Reproduce the journey yourself
I opened the site fresh, went through the full purchase journey, and watched what fired using Meta Pixel Helper and the browser console. The PageView fired. ViewContent fired. And then — the key events downstream weren’t firing reliably. Sometimes yes, sometimes no. Intermittent failures are the worst kind, because a single quick test can show “all good.”
Step 3: Into Google Tag Manager
The pixel ran through GTM, so that’s where the real detective work happened. GTM’s Preview mode shows exactly which tags fire on which triggers, in what order. Three things surfaced:
- A trigger tied to an element that had changed. A theme update had altered the page structure, and a trigger relying on it now matched only some of the time.
- Timing issues. On slower loads, the purchase event was racing the redirect — sometimes the browser navigated away before the tag finished.
- A duplicate, half-configured tag from an old setup, occasionally double-firing and muddying the data further.
None of these threw an error. All of them corrupted attribution.
Step 4: Fix, then verify like a sceptic
The fix itself was straightforward once diagnosed: rebuilt the trigger on a stable identifier, moved the critical event to a more reliable firing point, removed the zombie tag. Then verification — GTM preview, Pixel Helper, Meta’s Test Events tool, and most importantly, watching backend orders vs reported conversions re-converge over the following days.
The checklist that would have caught it early
- Compare platform-reported conversions vs backend orders weekly — a widening gap is your fire alarm.
- After any theme or site update, re-test the full event chain. Site changes are the number one silent killer of tracking.
- Keep GTM clean. Old tags and unused triggers aren’t harmless clutter — they’re future bugs.
- Never judge tracking by one test. Intermittent failures need multiple runs on different devices and speeds.
Ad budgets are decided by data. If the data is broken, every “optimisation” is just confident guessing. Audit your tracking like money depends on it — because it does.