Guide · Measurement
The tag is dead and the account doesn't know it
Written 5 March 2026 by Lena Verhoeven · checked 27 July 2026
A broken conversion tag is the quietest expensive thing in paid search: the site sells normally, the money lands normally, and the ad account slowly loses its mind. This guide explains why tags die, what automated bidding does during the silence, and the detection and recovery order that decides whether the incident costs three figures or five.
Why is this failure so quiet?
Because every system involved keeps functioning. The website works, orders arrive, the shop backend counts revenue, and Google Ads dutifully reports what it sees: conversions approaching zero. No error appears anywhere, since nothing is erroring; a tag that never fires is indistinguishable, from the platform's side, from customers who stopped buying. The founding incident of this company ran 62 hours on exactly that ambiguity, and the full arithmetic lives in the €9,400 write-up.
What are the usual causes of death?
| Cause | Typical trigger | Tell |
|---|---|---|
| Site releases | Theme or checkout deploy removes the snippet or breaks the trigger | Conversions stop at deploy time exactly |
| Tag-manager edits | Container change breaks a trigger condition | Some events die, others survive |
| Consent misconfiguration | Banner swap blocks tags for everyone; EEA accounts under the v2 regime are extra exposed | Volume drops by roughly the consent-decline share, or to zero |
| Platform migrations | Replatform ships without the measurement plan | Everything dies on launch day, amid celebration |
| Duplicate-fire fixes gone wrong | Deduplication logic overcorrects | Volume halves rather than vanishes, the sneakiest variant |
What does the silence cost?
Three invoices arrive. The direct one: spend continues at plan while bidding flies blind. The compounding one: tCPA and tROAS strategies interpret the famine as collapse and drift toward whatever noise still registers, so traffic quality degrades within days. The lingering one: after the fix, the blackout sits in the learning history, and bidding that is allowed to average it in stays skittish for weeks. On a mid-size account paying near the 2026 median of $66.69 per lead, even a two-day blackout's three invoices together usually clear four figures, which is why the heartbeat check earns its keep on the first catch.
How does detection actually work?
One comparison, run continuously: conversions in the recent window against a rolling same-weekday baseline, with silence beyond a threshold treated as an incident rather than a trend. The window must respect your volume: a store converting hourly can alarm after three quiet hours, while a B2B account with 20 monthly leads needs a day-scale window or it cries wolf weekly. Pair the heartbeat with a deploy-time correlation, since releases cause most deaths: our alerts print the last deploy timestamp next to the silence, which usually ends the diagnosis before it starts. The DIY version is an Ads script and an afternoon, per the monitoring checklist; the service version runs every 30 minutes under PPC monitoring.
The recovery order that limits damage
- Confirm and scope: which events died, when, and what still fires. Partial deaths mislead more than total ones.
- Fix the counting before touching campaigns; every judgment during the blackout is suspect.
- Fence the bidding: on thin-volume accounts, consider stepping tCPA toward manual control during the gap rather than letting it learn from famine.
- Re-base after the fix: reset targets from clean post-recovery data instead of averaging the scar into history.
- Write the incident down: timestamps, cost, cause, and the check that will catch the recurrence. The second €9,400 weekend cost €130 precisely because the first one got documented.
Asked before signing, usually
By default, you wouldn't; that is the whole problem. The site keeps selling, revenue keeps landing in your shop system, and Google Ads simply reports a store that stopped converting. Detection requires comparing conversion volume against a baseline and treating silence as a signal, which nothing does unless you build or buy it.
Site releases, by a distance: theme updates, checkout changes, tag-manager container edits, and consent-banner swaps. Then platform migrations, plugin updates on CMS stores, and consent misconfigurations that block tags for everyone instead of respecting choices. Almost never Google; almost always your own deploy pipeline.
It responds to the famine. Target-CPA and target-ROAS strategies read the silence as collapsing performance and start chasing whatever still registers, drifting spend toward junk-adjacent traffic within days. The damage therefore compounds: wasted spend during the outage plus poisoned learning after it.
Counting resumes immediately; trust takes longer. Bidding re-stabilizes over one to two weeks depending on volume, faster when you re-base targets on clean post-fix data instead of letting the algorithm average the blackout into its history. The re-basing step is the one most self-managed recoveries skip.
It adds a second failure surface. Since March 2024 EEA and UK accounts need v2 wired correctly, and a misconfigured banner that blocks tags outright looks identical to a dead tag in the data: silence. The heartbeat check catches both; only the diagnosis differs, which is why our alerts name the likely suspect.
When did your tag last prove it's alive?
If the answer involves checking, the heartbeat check pays for itself on its first catch.
Get a free strategy call