Operations
Part of Ecommerce analytics: from events to defensible decisions
Common ecommerce analytics questions answered, definitions first
Answers to six common ecommerce analytics questions, including why Klaviyo, Shopify and Google Analytics 4 revenue differ and how to reconcile them, with the definition behind every metric.
An illustrative US retailer brings a Klaviyo revenue figure to a marketing review beside Shopify Analytics and Google Analytics 4. The totals do not agree, and the team must decide whether the gap reflects different definitions, an incomplete period, or a reporting fault. Rather than nominate a winning dashboard, the analyst follows one reproducible case.
Six common questions, answered
Why do Klaviyo, Shopify Analytics and Google Analytics 4 show different revenue?
How do I reconcile the three totals?
What is a conversion rate?
What is the difference between repeat purchase rate and retention rate?
How should average order value treat discounts and refunds?
Does attributed revenue prove marketing caused the sale?
The sections below show the full case behind these answers.
Start with the decision, not the dashboard
The analyst starts by identifying what decision the revenue comparison is meant to support. That purpose determines which measures need to be compared and prevents a number from being treated as meaningful before its population, calculation, period, and limits are understood.
- Situation:The team brings the Klaviyo figure to a marketing review and sets it beside the totals in Shopify Analytics and Google Analytics 4. The reports appear inconsistent, so the team needs to establish what the difference can support.
- Checked:The analyst records each measure's population, unit, source, calculation, period, and order status. For attributed revenue, the record also states the model and lookback window, such as last click or time decay.
- Found:The tools answer different questions. Google Analytics 4 reports users and sessions, Shopify Analytics reports buyers and orders, and Klaviyo attributes revenue to flows and campaigns. Attribution assigns credit under a model; it does not establish incrementality.
- Done:The team chooses the definition that fits its decision, names an owner, and records what the result cannot establish. It marks unfinished periods provisional instead of refreshing reports until one produces a preferred answer.
The outcome is a decision record rather than a winner among dashboards. An observed relationship can guide research, but that relationship alone does not show that a campaign caused a purchase. The record keeps the evidence period, calculation, and known limits beside the result for the next reviewer.
Put the reports on one evidence sheet
The comparison becomes interpretable when every figure is described using the same set of questions. The analyst writes down each report's source, unit, population, period, calculation, and order status before comparing totals, then notes where the answers differ or remain unknown.
The evidence sheet distinguishes dimensions, which describe attributes, from metrics, which measure quantities. It also records platform scope and compatibility limits. A controlled dictionary gives the retailer a place to maintain its own formulas, states, and source systems instead of allowing interface labels to become business definitions by default.
| Measure | Definition to record |
|---|---|
| Conversion rate | Orders divided by sessions, for a stated storefront and period |
| Average order value | Revenue divided by orders, with refunds, discounts, and shipping treatment stated |
| Repeat purchase rate | Share of customers with more than one completed order |
| Retention rate | Share of a starting cohort that orders again in a later period |
| Customer lifetime value | Expected gross profit per customer across the relationship, not one order |
| Customer acquisition cost | Acquisition spend divided by new customers |
| ROAS and MER | Ad-attributed revenue divided by ad spend; total revenue divided by total marketing spend |
These definitions make the comparison more reproducible, but they do not make dissimilar reports equivalent. For a public reference, the Census Bureau's ecommerce reporting language and Eurostat's metadata for online purchase statistics provide statistical context; neither defines the retailer's own order status, customer identity, or revenue calculation.
Dimension vs Metric
Dimension
- What it is
- describes attributes
- Example
- user, event, item
- Compatibility
- limited pairings
- Ownership
- named steward
Metric
- What it is
- quantitative measurement
- Example
- counts, totals, averages
- Compatibility
- limited pairings
- Ownership
- named business owner
The analyst uses the same evidence sheet to decide whether a field belongs in the report. A standard field remains the default when its documented meaning fits the decision. A custom field needs an approved purpose, bounded options, clear scope, and a named owner before it is added.
Custom Field Justification
- Approved decision identified
- Bounded values defined
- Clear scope set
- Named owner assigned
- No IDs or unrestricted text
- Metric card recorded
Before finalising a new field, the analyst checks whether it applies to users, events, or items and whether the platform supports that scope. Customer IDs, order IDs, timestamps, email addresses, and unrestricted text are rejected as ordinary reporting dimensions, even if a form accepts them, because variation can weaken reports and raise privacy and access risk.
Interpret the gap without claiming a cause
The finding is that the three reports are not direct substitutes. Google Analytics 4 describes users and sessions, Shopify Analytics describes buyers and orders, and Klaviyo assigns revenue to flows and campaigns. A revenue attribution result describes assigned credit under chosen rules, not proof that marketing created additional sales.
The analyst records the attribution method and lookback window beside the credited revenue. First click, linear, position based, time decay, data driven, and last click are available approaches, but none supplies a neutral account of sales. Each method assigns credit according to its own rules, so incrementality needs a separate basis.
The record also checks whether the period is mature enough for the decision. Late events, refunds, returns, renewals, chargebacks, and cohort age can change a result; the analyst notes the processing delay, last complete date, and known gaps. An unfinished period stays provisional instead of being presented as a settled comparison.
Close the case with a bounded action
The team selects the measure that fits its stated decision and records its limits, rather than trying to force all three systems to agree. The metric card preserves the question, calculation, sources, owner, period, known gaps, approved audience, privacy review, and next action so another analyst can retrace the reasoning.
The saved record also states data protection limits and how personal information is handled. A US retailer includes applicable legal review and platform data-processing terms; FTC security guidance is a reference point, not a substitute for that review. The record gives the team a basis to distinguish a changed definition from a changed customer pattern.
Rules carried into the next review
The reusable rules from the case are to define the measure, state its limits, mark incomplete periods, and preserve the decision record. Together, these steps let a later reviewer check whether the source, population, or calculation changed before concluding that customer behaviour shifted.
Keep fields only when their definitions serve an approved decision, and name an owner for each. Behavioural data can locate patterns, while research and operational evidence are usually needed to explain why customers act. Close reports and rejected options when the decision cycle ends, then set a date for the next review.
