Analytics architecture · GTM · GA4 · Paid media · BigQuery

Multi-Domain Tracking & Conversion Measurement Rebuild

The business had traffic, leads and media spend — but no trustworthy conversion number. Old tags still fired, the same form submission reached reporting twice, and regional websites used different event logic. I rebuilt the measurement system so GA4, Google Ads, Meta and CRO teams worked from one governed signal.

Company

Northline Group · anonymized

Scope

12 regional domains

My role

Senior Digital Analyst

Core outcome

One trusted conversion system

Abstract analytics data streams converging into one governed measurement hub — Measurement architecture12 domains converging into one trusted conversion signal

Measurement control · after rebuildProduction verified

Sources12 regional domainshostname + location_id

Single sourceGTMgoverned container

AnalyticsGA4one property

Paid searchGoogle Adsprimary actions

Paid socialMetaverified event

WarehouseBigQueryQA layer

Anonymized architecture · no production IDs0 duplicate routes

Tags audited47

Every tag, trigger and platform connection inventoried.

Legacy tags retired19

Paused, quarantined, approved and removed.

Domains standardized12 / 12

One container, property and event taxonomy.

Duplicate key events0.4%

Down from an 18.7% audit baseline.

The business problem

The company could not tell whether performance changed — or tracking changed.

GA4, Google Ads and Meta reported different lead totals. Some domains sent two page views per load; a form click could be counted even when validation failed; and historic tags still passed conversions to ad platforms. Channel ROI, cost per lead and site-to-site comparisons were therefore unreliable.

That also blocked conversion optimization: an A/B test cannot be trusted when its primary conversion fires twice or fires before the user has actually completed the action.

My Role

I owned the end-to-end measurement rebuild: diagnosis, stakeholder decisions, implementation, cross-platform QA, documentation and the reporting layer used to monitor the new setup.

Export and classify the complete GTM inventory
Agree Keep / Fix / Stop / Remove decisions with PPC, product and development
Design the multi-domain event and parameter taxonomy
Configure GA4, Google Ads and Meta delivery through GTM
Build BigQuery QA logic and the stakeholder dashboard

Applied Workflow

01

Inventory

Tags, triggers, variables, scripts and account links.

02

Reproduce

Test every hostname, device and conversion path.

03

Decide

Approve ownership and retirement with stakeholders.

04

Rebuild

One container, taxonomy and source of truth.

05

Activate

Verified signals to Ads, Meta and GA4.

06

Monitor

BigQuery reconciliation and alert thresholds.

Google Tag ManagerInventory, folders, triggers and release control

Google Analytics 4Event model, key events and DebugView

BigQuery + SQLDeduplication, hostname views and QA

Google AdsPrimary / Secondary conversion governance

Meta Events ManagerDataset, custom events and diagnostics

Tag AssistantNetwork-level firing validation

Looker StudioAcquisition and measurement health dashboard

Tracking PassportLiving ownership and change-control document

i

Portfolio-safe case. The company, domains, IDs and absolute volumes are fictionalized. Counts and percentages are normalized representative data; the audit logic, architecture, governance process and implementation decisions reflect the real work.

01 · Audit before changing production

Every tag received an owner, a purpose and a decision.

I did not delete unknown tags on sight. First I traced what each one sent, reproduced its trigger, checked the destination account and documented the business action. The resulting table was reviewed with PPC, product and development before anything was paused.

Keep12

Correct, required and owned.

Fix / rename8

Valid purpose, broken logic or naming.

Stop19

Legacy, duplicate or no longer used.

Remove8

Obsolete scripts after observation window.

GTM audit registerAnonymized reconstruction

Tag / triggerObserved behaviorDecisionReason and action

GA4 · Google tag · All pagesInitialization — All Pages

Two configuration calls on 4 domainsFixKeep one GTM implementation; remove the CMS copy.

UA · All Pages · LegacyPage View — All Pages

Still sending sessions to an unused propertyStopPause, move to legacy folder, remove after 14-day QA.

GA4 · lead_click_oldClick — CSS selector

Fired before form validationStopReplace with success callback and unique event_id.

Ads · Lead import from GA4Imported key event

Same outcome also sent directly from GTMSecondaryRetain one bidding route; keep import for comparison only.

Meta · Base pixel · CMSCustom script

Loaded alongside the GTM base pixelRemoveCentralize ownership in GTM and verify Test Events.

GA4 · pricing_viewCustom event — pricing module

One event with valid hostname and location_idKeepStandardize name and document as a micro-conversion.

01Evidence

Double Google tag

Tag Assistant and the network log showed two GA4 configuration requests on four sites: one from the CMS and one from GTM.

Remove the CMS script and make GTM the sole owner.

02Evidence

False-positive lead

The click trigger fired even when required fields failed validation. A second event fired after the successful server response.

Trigger only on confirmed success; attach event_id.

03Evidence

Duplicate Ads route

The same lead reached Google Ads from a direct GTM tag and an imported GA4 key event, both marked Primary.

Use one Primary route and demote the comparison signal.

Cleanup followed a controlled retirement process

Approved legacy tags were first paused and moved into 99_Legacy_Stopped with date, owner and replacement noted. After a 14-day observation window and regression QA across all hostnames, obsolete tags were deleted and the decision was recorded in the Tracking Passport change log.

01Stakeholder approval

02Quarantine folder

03Observation + removal

02 · Rebuild the measurement architecture

One container and one GA4 property — without losing local detail.

Twelve regional websites needed portfolio reporting and site-level accountability. I deployed the same GTM container and GA4 stream across every domain, then used hostname, location_id and form_type to separate performance in BigQuery and reporting.

Production event flow

12 / 12 domains covered

domain-a.testpage_hostname = east

domain-b.testpage_hostname = west

+ 10 hostnameslocation_id required

GTMshared container
versioned release

GA4behavior + key events

Google Adsapproved conversions

Metacustom lead event

BigQueryraw QA + reporting mart

Business event taxonomySame name and definition across every hostname

Event	Trigger condition	Required parameters	GA4	Ads use	Meta use
lead_submit	Confirmed form success response	event_id, hostname, location_id, form_type	Key event	Primary	Optimize
schedule_complete	Booking confirmation received	event_id, hostname, location_id	Key event	Primary	Optimize
phone_click	Tap on tracked phone link	hostname, location_id, placement	Key event	Secondary	Measure
contact_start	First valid form interaction	hostname, location_id, form_type	Standard	Secondary	Measure
pricing_view	Pricing module becomes visible	hostname, location_id, page_type	Standard	Observe	Observe

-- Daily duplicate and hostname QA
SELECT
  event_date,
  REGEXP_EXTRACT(page_location, r'^https?://([^/]+)') AS hostname,
  event_name,
  COUNT(*) AS event_rows,
  COUNT(DISTINCT event_id) AS unique_actions,
  SAFE_DIVIDE(event_rows - unique_actions, event_rows) AS duplicate_rate
FROM analytics.events_*
WHERE event_name IN ('lead_submit','schedule_complete')
GROUP BY 1,2,3;

Why BigQuery was necessary

GA4 interface totals alone could not expose the same event_id arriving twice or identify which hostname lost tracking after a release.

✓Deduplicate confirmed actions by event_id.
✓Compare domains through hostname and location_id.
✓Reconcile paid-platform totals with validated leads.
✓Alert when event volume or parameter coverage changes.

Governance deliverable

Tracking Passport

A living source of truth for Analytics, PPC, CRO and Development — updated with the implementation, not after it.

Open the template↗

The passport lets a new analyst understand what is connected, what fires, where each event goes and which actions influence bidding — without opening every platform first.

01Accounts & identifiersGA4, GTM, Google Ads, Meta and supporting platform IDs.

02Platform connectionsGTM → GA4 / Ads / Meta and GA4 ↔ Ads status with QA notes.

03GTM inventoryExact tag, trigger, status, purpose and Keep / Fix / Stop decision.

04GA4 event mappingBusiness action → dataLayer event → trigger → tag → key-event status.

05Ads conversion reviewPrimary vs Secondary decision and the single approved delivery route.

06Change log & handoverDate, component, reason, owner and production validation.

03 · Activate and verify paid-media signals

Ads and Meta received the same confirmed business outcome — once.

The rebuild separated optimization outcomes from diagnostic micro-actions. Google Ads used a single Primary route for each approved conversion, while Meta received a custom form-completion event from the same confirmed success signal.

Conversion governance after rebuildExact IDs redacted · representative action names

Platform	Action	Source	Optimization	Status	QA decision
Google Ads	Form Submitted	Website · GTM	Primary	Active	Confirmed success only; unique event_id
Google Ads	Booking Completed	Website · GTM	Primary	Active	Confirmation callback; cross-domain linker verified
Google Ads	Contact Started	Website · GTM	Secondary	Active	Observation only; excluded from bidding
Google Ads	GA4 Lead Import	GA4 import	Secondary	Comparison	Prevent duplicate optimization route
Meta	WebsiteFormSubmitted	Dataset · GTM	Optimize	Receiving	Custom event validated in Test Events
Meta	ContactStart	Dataset · GTM	Measure	Receiving	Funnel diagnostic; not the primary outcome

Tag Assistant✓

One network request

Each approved event and platform destination fired once after a confirmed action.

12 hostnames · desktop + mobile

GA4 DebugView✓

Parameters present

hostname, location_id, form_type and event_id arrived with the expected values.

Schema coverage: 99.6%

Google Ads✓

Bidding signal clean

Only verified business outcomes remained Primary; imports were Secondary.

Diagnostics: recording conversions

Meta Test Events✓

Custom event received

WebsiteFormSubmitted appeared once with the correct source URL and location key.

Dataset status: active

Acquisition & tracking health dashboard

Data fresh · 52 min

Sessions124.9Kall hostnames

Validated leads1,855unique event_id

Ads ↔ GA4 delta4.8%inside 10% threshold

Duplicate rate0.4%inside 1% threshold

Validated leads by acquisition channelNormalized 30-day view

Google Ads

812

The rebuild corrected reporting and made optimization decisions defensible.

The immediate outcome was not “more conversions” created by tracking. It was a reliable definition of conversion, a clean signal for bidding, faster issue detection and a measurement layer that CRO could safely use for baselines and experiments.

Duplicate key-event rate18.7% → 0.4%

Repeated confirmed actions removed through trigger fixes and event_id QA.

Six-week production readout

Ads vs validated lead delta31% → 4.8%

One bidding route and scheduled reconciliation replaced mixed totals.

Inside agreed 10% tolerance

Common event schema4 / 12 → 12 / 12

Every domain now uses the same definitions and required parameters.

Hostname-level visibility retained

Issue investigation1–2 days → <30 min

Passport ownership and dashboard thresholds exposed the failing layer.

Typical tracking incident

−26%reported Ads conversions

The first “drop” was the proof that cleanup worked.

After duplicate and micro-conversion actions were removed from Primary reporting, Google Ads showed 26% fewer conversions while validated lead volume stayed stable. This was not lost demand. It was inflated measurement being corrected, so bidding could optimize toward real outcomes instead of easy-to-trigger events.

Practical business valueWhat teams could do differently after launch

01

Paid media optimized to real outcomes

PPC could distinguish Primary business conversions from diagnostic micro-actions and compare channel efficiency without double counting.

02

Portfolio performance became comparable

Leadership could view acquisition, conversion and data quality by hostname while keeping one governed GA4 property.

03

CRO gained a trustworthy primary metric

Experiments could use confirmed lead_submit or schedule_complete events, deduplicated in BigQuery and segmented by domain, source and device.

01

Reliable baselines

Form completion, channel CVR and hostname benchmarks no longer changed because of duplicate tags.

02

Experiment readiness

Exposure, variant, event_id and outcome could be joined in BigQuery for valid A/B-test analysis.

03

Faster diagnosis

A performance decline could be separated from a missing tag, parameter break or source attribution issue.

04

Governed change

Every new tag or conversion required an owner, passport entry, QA evidence and release note.

Why this matters for conversion optimization

experiment_exposurevariant_idlead_submitschedule_completeevent_idpage_hostnamesource_mediumdevice_category

Without trustworthy events, CRO can optimize to a tracking artifact. The rebuild established the measurement contract required to identify friction, prioritize hypotheses and calculate real experiment impact.

What this case demonstrates

Measurement architectureDesigned one scalable GTM and GA4 system across a multi-domain portfolio.

Governance & stakeholder alignmentTurned an unknown legacy container into approved ownership and change control.

Paid-media activationConnected verified conversion outcomes to Google Ads and Meta without duplicate routes.

BigQuery + CRO enablementBuilt QA, reconciliation and experiment-ready data instead of another passive report.

Other Case Studies

Product analyticsE-commerce Funnel Analytics→Lead analyticsLead Qualification Strategy→Measurement systemsBack to Top↑