GA4 architecture
GA4 is built on an event model. Every user interaction is an event with parameters:
event_name: "purchase"
parameters:
transaction_id: "T123"
value: 99.90
currency: "USD"
items: [{item_id: "SKU123", item_name: "...", quantity: 1, price: 99.90}]
Unlike Universal Analytics, GA4 has no special page-view hit. page_view is simply an event like any other.
Key concepts
| Concept | GA4 | Universal Analytics (retired) |
|---|---|---|
| Unit of data | Event | Hit (pageview, event, transaction) |
| Identifier | User-ID + Client-ID + ML modelling | Client-ID (cookie) |
| Funnels | Explorations → funnel report | Goals (fixed) |
| Sampling | None when integrated with BigQuery | Applied on high traffic |
| Web + app | One property | Separate accounts |
E-commerce in GA4
GA4 ships with a built-in event schema for e-commerce. The essential events:
view_item — a PDP view
add_to_cart — adding to the cart
begin_checkout — starting checkout
purchase — a completed purchase
Additional events for extended analytics:
view_item_list — products shown in a listing
select_item — a click on a product
view_promotion — a banner or promotion shown
select_promotion — a click on a banner
Correctly implemented GA4 e-commerce tracking gives you the purchase funnel, product reports, promotion performance and traffic source reporting out of the box.
Tip: if you run a personalization platform — recommendations, popups — make sure
view_promotionandselect_promotionfire when widgets are shown and clicked. That is what lets you attribute their contribution to revenue inside GA4.
The limits of GA4
GA4 does not replace specialised analytical tools:
– For product analytics (Amplitude, Mixpanel), GA4 is less flexible in ad-hoc segmentation
– For A/B testing, GA4 is an observation tool, not a decision-making one
– For technical diagnostics — latency, error rates — you need an APM, not GA4
GA4 is the standard for marketing analytics, attribution and SEO. For product analytics and experimentation it is used alongside specialised tools.