Owner, trigger and scope
The analytics owner is the incident lead: they confirm the outage, coordinate the response and close the incident. The developer owns fixes to site code, the data layer and server-side endpoints. The paid media manager owns protective action in the ad platforms. Each role has a named deputy, because outages rarely happen at convenient times.
The trigger is any alert or anomaly suggesting events or conversions have stopped or changed: an automated alert, a platform diagnostic warning, a flat line on a dashboard, or someone noticing that the ad platform and the CRM disagree far more than usual. Anyone may raise a suspected outage; only the analytics owner declares one.
The scope covers loss or corruption of measurement in analytics, tag management and ad platform conversion tracking, including browser and server-side paths. It does not cover site downtime, which follows the technical team's own incident process, though the two often overlap. Routine data quality checks are covered elsewhere; this SOP is for when something has visibly broken. When the email platform, form tool or booking tool itself is unavailable on the supplier's side, rather than the tags that measure it, the guide to preparing for a marketing tool outage covers the fallbacks, the pauses and the catch-up.
Before you start: inputs and access
| Input | Owner | Why |
|---|---|---|
| Incident log template | Analytics owner | Records times, decisions and evidence |
| Measurement plan | Analytics owner | Lists what should be recording and where |
| Container and release history | Developer | Shows what changed and when |
| A source outside the tags | Finance or CRM | Orders or leads to compare against |
| Ad platform access with edit rights | Paid media manager | Needed for protective steps |
| Stakeholder contact list | Analytics owner | Who to tell, in what order |
Make sure all three roles have the access they need before any outage happens. Discovering that only one person can publish the GTMTag management system. container, and they are on leave, is a common way an outage stretches from hours into days.
The procedure
Confirm the outage
Compare the suspect measurement against a source outside your tags, such as orders in the shop platform or leads in the CRM, for the same period. If the business system shows normal activity while the measurement shows a drop, treat it as an outage. Note the first time the data looks wrong, open an incident log and tell the developer and paid media manager that an incident is open.
Triage the cause
Check what changed near the start time: site releases, container versions, consent bannerSoftware that collects, stores and passes on visitors' consent choices for cookies and tracking. updates, payment or form provider changes, and platform notices. Use preview and debug tools, described in the debugging and preview guide, to watch whether the event fires and whether the data reaches GA4Google's event-based analytics platform. and each ad platform. Narrow the fault to one layer: page, tag, consent, browser, server or platform.
Assess impact on bidding and reports
List which conversion actions are affected, on which platforms, since when, and which campaigns bid on them. Note whether server-side events through the Conversions APIServer-side event sharing with Meta. are still arriving, since one path may cover for the other. List the reports and dashboards that will show the gap. Write this in the incident log; it is the basis for every later decision and update.
Protect campaigns
The paid media manager decides how to stop automated bidding reacting to the missing signal. Typical options are holding budgets and targets steady, avoiding structural edits, and, where the platform documents it, using a data exclusion for the outage windowA period when conversion or analytics tracking stops recording, or records wrongly, while sales carry on; its start and end are marked as the outage window.. Pausing is a last resort for campaigns that are clearly overspending on bad signals. Record every action and its time in the incident log.
Fix and deploy
The developer or analytics owner makes the smallest change that restores tracking: rolling backReturning a changed system, such as a tag container, workflow or page template, to its last working version, planned and rehearsed before the change goes live. a container version, fixing a selector, restoring a data layer push or correcting a server endpoint. The conversion tracking setup guide and the Conversions API guide describe how each should be configured. Test in preview, then deploy through the normal release path with a note referencing the incident.
Verify recovery
Generate test conversions on the live site and confirm they appear in analytics and in each ad platform's diagnostics. Then watch real data for long enough to see normal volumes return and compare again with the business system. Recovery is confirmed only when real conversions, not just test ones, are recording at their usual relationship to orders or leads.
Annotate and report
Add an annotation to analytics and the reporting dashboards marking the outage window. Write a short incident report: what broke, when, the impact, what was done and what will stop it recurring. Upload offline conversions where the platform supports it and the business has the records. Share the report with the stakeholders and close the incident.
A consent banner update that stopped purchase tracking
Kettleby Coffee, an invented coffee subscription business, notices on a Thursday morning that analytics purchases have fallen to 3 for the previous day against 140 orders in the shop. Triage finds a consent banner update released on Wednesday at 16:00 that delays the purchase tag until after the confirmation page unloads. The paid media manager holds budgets and applies a data exclusion for the window; the developer rolls back the banner configuration by 11:30. Real purchases match orders again by mid-afternoon, 96 orders are uploaded as offline conversions, and the incident report recommends a tracking check in every banner release.
QA checks
A second person, usually the paid media manager, should read the incident report before closure and confirm it matches what they saw.
Before the incident is closed, the analytics owner confirms:
- Outage window is identified with a start and end time and evidence for both.
- Root cause is identified, not just the symptom.
- Fix is deployed through the normal release path and documented.
- Real conversions are recording at their usual relationship to business records.
- Every affected platform shows healthy diagnostics, including server-side paths.
- Protective actions are reversed where they were temporary, such as data exclusions or held changes.
- Annotations are in place in analytics and in reporting dashboards.
- Incident report is written and shared.
Hand-offs and records
During the incident, the analytics owner sends short updates to stakeholders at agreed points: when the outage is confirmed, when the cause is found and when recovery is verified. The paid media manager hands any campaigns that were adjusted back to normal optimisation once recovery is confirmed, with a note in the pause log. The marketing analyst receives the outage window for the next monthly pack. Where a supplier or agency caused the outage, send them the report too. The academy exercise on handling a tracking outage and a wrong report for a stationery brand practises running an outage from detection to annotation when a report built on the missing week has already reached the board.
The incident log and report are the lasting records. File them where future incidents can be compared. Record the outage window in the shared measurement change log as well, so that it sits beside every other change when someone later explains a step in the data. A short session using the five whys method for root-cause analysis is a practical way to trace an outage from the broken tag to the missing check that let it through. Recurring causes, such as releases that ship without a tracking check, should become changes to the release process; the data quality guide covers the routine monitoring that catches the next outage sooner. When the error is in AI-assisted output customers saw rather than in tracking, the guide to handling an AI mistake follows the same contain, correct and trace sequence. After each incident, update the tracking row of the marketing risk register: whether the warning sign appeared in time, whether this procedure worked and whether the risk's likelihood should be rescored.
Common mistakes
Outages are stressful, and most of these mistakes are shortcuts taken to end the stress quickly.
- Declaring an outage from one dashboard. Compare with a source outside the tags first.
- Pausing everything. Pausing loses delivery and learning. Protect bidding more precisely.
- Fixing in production without testing. A rushed fix can break a second event. Use preview.
- Verifying with test events only. Real conversions confirm recovery; test events confirm a path.
- Forgetting the server-side path. Browser and server events can fail separately. Check both.
- No annotation. The gap reappears unexplained in every year-on-year comparison.
- No follow-up change. The same cause returns at the next release.
- Silent incidents. Stakeholders who hear late lose trust in every number. Send updates early.
