THE SHORT ANSWER
Use a hierarchy that connects demand to paid orders, customer behaviour and contribution. Report sessions or users with their definitions; conversion, AOV, CAC and ROAS with their assumptions; retention and LTV by cohort; and profitability with the included costs visible.
Organize metrics around the business system
| Layer | Core questions | Example measures |
|---|---|---|
| Acquisition | Who arrived and at what cost? | Qualified sessions/users, new customers, CAC, channel mix |
| Conversion | Where did suitable demand progress or stop? | Product views, add-to-cart, checkout and paid-order conversion |
| Order economics | What did the completed order create? | AOV, items per order, discount, gross margin, contribution |
| Retention | Did customers return at a relevant time? | Repeat rate, frequency, cohort retention, realized value |
| Profitability | Does growth create value after relevant costs? | Contribution, payback, return cost and operating profit context |
Evidence & context: Google Analytics Help · Google Analytics Help
Definitions are part of the metric
Conversion rate can use sessions, users or eligible product-page visits as its denominator. AOV can include or exclude tax, shipping and refunds. CAC can be channel-assigned or blended. ROAS depends on an attribution model. None has one universal implementation.
Maintain a metric dictionary with owner, formula, source, scope, exclusions, latency and known limitations. When a platform changes a definition, preserve the date so apparent trends are not mistaken for customer behaviour.
Evidence & context: Google Analytics Help · Google Analytics
A vanity metric lacks a decision context
Traffic, followers and impressions are not inherently useless. They become vanity measures when presented as success without the quality, cost or downstream outcome needed for a decision.
| Observation | Diagnostic addition | Possible decision |
|---|---|---|
| Traffic increased | Source, landing page, new customers and conversion | Scale, repair message match or stop low-quality activity |
| ROAS improved | Attribution, margin, incrementality and customer mix | Reallocate only if business value also improves |
| AOV fell | Product mix, quantity, discounts and conversion | Change merchandising or accept a healthy mix shift |
| Repeat rate fell | Cohort maturity, product cycle and experience | Fix service, timing or segment expectations |
The average can hide the operating problem
Segment by new versus returning customer, product or category, device, geography, acquisition source, promotion and cohort. Stop when the segment becomes too small or unstable to support the claimed precision.
Use trends and comparable periods, then annotate launches, stock-outs, tracking changes and promotions. A dashboard without operating context can turn a known event into a false mystery.
Use calculators as transparent working models
The OpenSkool ROAS Calculator and Acquisition Cost Calculator make their formulas visible. Use them to explore assumptions, not to replace reconciled finance and analytics data. Then build a growth dashboard around decisions.
Sources & further reading
- Ecommerce in Google Analytics
Google Analytics Help. Official documentation for ecommerce events and reports. A measurement implementation does not by itself establish causality or profitability.
- Ecommerce purchases report
Google Analytics Help. Official definitions for item-level commerce metrics. Revenue fields have different inclusions, so teams must document the field they use.
- Understand user metrics
Google Analytics Help. Official definitions for total, active, new and returning users. Identity limits and configuration can affect interpretation.
- API dimensions and metrics
Google Analytics. Official GA4 reporting definitions checked 13 September 2026, including session source, medium, referral and landing-page dimensions. Attribution remains limited to observable interactions.
Examples and exercises are illustrative unless attributed to a source. No independent expert review is claimed.
A correction, a counterexample or an experience worth sharing?
Join the conversation ↗