GA4 does not know order economics
Refunds, cancellations, discounts, product cost and margin live in the store or ERP.
BigQuery · Google Cloud · GA4 · Google Ads
BigQuery and GCP as a shared data layer for marketing, e‑commerce and SaaS.
I connect raw GA4 events, Google Ads cost and campaign structures with store, CRM, product and billing data. The result is a stable foundation for reporting, customer analysis, retention, LTV and further modelling.
GA4 export is an entry point, not a finished solution. Value appears after definitions are agreed, data is transformed and tested, and purpose-built analytical layers are created.
GA4, Ads and business datashared identifiers, calendar and definitions
Analysis-ready modelse-commerce, SaaS, customer, campaign and product
Quality, cost and ownershiptests, monitoring, IAM and query controls
Starting point
When reports from individual tools are no longer enough.
BigQuery makes sense when a business question requires connecting user behaviour with acquisition cost, orders, margin, customer status or product usage.
SaaS analysis ends at registration
Trial, activation, plan, payment, MRR and churn are split between product, CRM and billing.
Campaign cost remains isolated
Advertising platforms use different structures, windows, currencies and aggregation levels.
No shared customer view
Sessions, leads, transactions and subscriptions lack stable keys and identity rules.
Reports depend on manual exports
Spreadsheets and dashboards replicate logic but produce different, hard-to-audit results.
Models lack reliable inputs
LTV, churn, propensity and forecasts require historical, documented and controlled features.
Product architecture
From sources to models that answer business questions.
Raw data is separated from transformations and analytical marts, so dashboards, SQL analysis and ML use the same definitions.
01 · SourcesGA4, Google Ads, store, CRM, billing, productnative exports, Data Transfer Service, APIs, files or backend events
02 · Raw dataBigQuery raw and stagingoriginal tables, load history, schema and freshness controls
03 · TransformationsSQL and Dataformdependencies, incremental models, tests, documentation and schedules
04 · Data productsE-commerce, SaaS, marketing and customer martsBI, ad-hoc analysis, experiments, segments and BigQuery ML
One metric definitionlogic is not copied into every dashboard
Every model has testsuniqueness, completeness, freshness and expected ranges
Every query has a costpartitions, clusters, limits and smaller reporting tables
Three usage paths
A data layer matched to the business model.
E-commerce
- net revenue, refunds, cancellations and margin,
- funnels, products, baskets and promotions,
- CAC, ROAS and campaign profitability,
- cohorts, purchase frequency and LTV,
- customer segments and analysis recommendations.
SaaS
- acquisition, trial, activation and paid,
- MRR, ARR, expansion, contraction and churn,
- cohort retention and feature adoption,
- CAC payback and LTV by segment,
- product, CRM and billing integration.
Modelling
- sales, demand or revenue forecasts,
- churn and propensity with reliable labels,
- LTV estimation and segment scoring,
- datasets for attribution and experiments,
- BigQuery ML or models outside the warehouse.
Predictive modelling is optional. Data quality, stable definitions, sample size and a way to act on the result come first.
Implementation scope
From GCP design to usable data products.
The scope may stop at a reliable foundation or include complete models, dashboards and first analyses.
- 01Outcome: architecture and ownership
GCP design and security
Project, billing, region, datasets, service accounts, IAM, environments and access rules.
- 02Outcome: controlled raw data
Source integration
GA4 export, Google Ads transfer and agreed store, CRM, billing or product connections.
- 03Outcome: analytical tables
Transformations and data model
Staging, core and marts with keys, calendar, currency, sessions, customers and business events.
- 04Outcome: repeatable pipeline
Automation and quality
Dataform or an agreed workflow, dependencies, incremental models, assertion tests, schedules and alerts.
- 05Outcome: business use
BI, analysis and modelling
Tables for Looker Studio, Power BI or Tableau, SQL analysis and optional predictive models.
- 06Outcome: maintainable solution
Documentation and handover
Definitions, lineage, operating instructions, costs, known limits, owners and training.
Tangible outcome
An e-commerce model connects the click to the order result.
For SaaS, an equivalent data product connects acquisition with activation, subscription, revenue and churn.
Analytical martfct_order_performancee-commerce
Traffic and campaignGA4
source, medium, campaign, landing page, session and user identifiers.
Media costGoogle Ads
campaign ID, clicks, impressions, cost and agreed aggregation.
Business resultStore or ERP
order ID, status, revenue, discount, refund, product cost and margin.
Ready metricsBigQuery mart
net revenue, margin, ROAS, CAC, conversion rate and cohort LTV.
Acceptance criteria
The pipeline must be reliable, observable and economical.
Quality and freshness thresholds are defined for every source because marketing and business data update at different rhythms.
Honest limitations
BigQuery does not create value without good data and decisions.
- GA4 export is not a copy of GA4 reports.Raw events differ because of modelling, attribution and processing logic.
- Streaming does not guarantee completeness.Intraday is useful for fast views; stable reconciliation should use daily export.
- Google Ads has its own update rhythm.Refresh windows and backfills must account for data corrections.
- Customer identity needs rules.BigQuery will not automatically connect an anonymous session, lead, account and payment.
- ML does not replace a hypothesis.It needs a label, enough observations, validation and a plan for using the output.
- Cloud requires governance.IAM, billing, regions, retention, sensitive data and changes need owners.
Analytics and data engineering
I connect business requirements, SQL, GA4 and reporting.
I lead projects from metric definitions through data models and dashboard acceptance, keeping GCP architecture connected to e-commerce, SaaS, marketing and finance questions.
20+years of experience
150+analytics and data projects
50+clients
Figures refer to the broader analytics and data practice, not only BigQuery implementations.
Before we talk
Frequently asked questions.
Is GA4 export alone enough?
Not for recurring reports or analysis. Raw export needs session and event models, metric definitions, tests, cost controls and audience-specific tables.
Can Google Ads be integrated without a third-party tool?
Yes. BigQuery Data Transfer Service provides a Google Ads connector with account, table, schedule and refresh-window configuration.
Which sources should be added beyond GA4 and Ads?
Usually orders, refunds, product cost, CRM, leads, subscriptions, payments and backend events — selected by the question, not integration count.
Does BigQuery suit e-commerce and SaaS?
Yes. E-commerce connects campaigns, orders, products and margin; SaaS connects acquisition, activation, usage, billing, MRR and churn.
Can the scope include BigQuery ML?
Yes, with a valid use case and sufficiently reliable data. Modelling is a separate module with validation and an action plan.
How are BigQuery costs controlled?
Through data modelling, partitioning, clustering, materialisation, query limits, budgets and monitoring. Dashboards should use reporting marts rather than raw events.
First step
Start with the sources and three decisions the data must support.
Describe the GA4 export, advertising accounts, store or billing platform and the questions that cannot currently be answered reliably.
Prepare the project