Start with one business question, then decide what is safe to change
13 parts · 2 checkpoints · 1 free / 12 member parts →
Data analysis begins by narrowing one business question. Keep the SKU, market, UTM, page version, dates, and cohort in view, and state what GA4, Shopify, ad platforms, Finance, and support can each prove. Then reconcile order identity, order and refund state, and the 7- and 14-day maturity windows. If source, scope, or refund state is unstable, freeze this week’s budget, page, SKU, or automation change.
The Google Analytics event and transaction_id guides and Shopify Acquisition report help check measurement boundaries. They cannot prove this store’s event quality, orders, refunds, attribution, profit, customer outcomes, or automation release. The pet rear-seat protector numbers below are a bounded teaching case.
What this lesson answers
- When a data anomaly appears, how do you define an answerable business question before changing budget or a page?
- How should one action record retain SKU, market, UTM, page version, dates, and cohort?
- Why should a row stay not yet trusted when transaction_id mismatches, duplicate firing, or refund maturity is unresolved?
- When an Acquisition report shows visitor source, how do you avoid treating platform credit as orders or profit?
- After reviewing an anomaly, how do you write sources, owner, action, rollback point, and review date into one weekly action row?
This week’s data decision
Analytics is not looking at more reports; it reduces wrong actions. GA4 measures site behavior, while Shopify records order, payment, and refund facts. Define the business question first, then evidence, source, responsible team, action, and review date so the records do not argue in separate languages.
Standalone case: In May 2026, an independent store selling pet-travel gear to US drivers is still validating a rear-seat protector; this week it changes no budget and asks only whether verified installation, size, and fit proof reduced confusion. The current anomaly is seven GA4 purchase events that deduplicate to five unique transaction_ids matching five unique Shopify orders, with one refund not yet mature. A UTM is a link-source tag, and a transaction_id is an order identity key; they find the same transaction but cannot alone prove ads, pages, or profit improved. Reconcile GA4/UTM/purchase transaction_id, Shopify order and refund state, support fit tags, and contribution profit per order; any mismatch is “not yet trusted” and freezes budget, SKU-scale, and profit conclusions. Use seven days for cart/purchase and 14 for refunds, support, and profit; these are only this sample-validation windows.
- Problem -> Evidence: Question first
- Result / Process / Diagnostic: Metric layers
- GA4 / Shopify / Ads / Finance: Source roles
- Responsible team + action + date: Review output
Reconcile identity and state first
GA4 shows seven purchase events, Shopify shows five unique orders, and one refund is still immature. Which action is supported by those facts?
What can these numbers support?
Self-check not answered
Write one trusted action row
Make question, source, identity, state, and maturity pass together before changing action.
Free account to read · Parts 2-13 · 12 parts
Members: full exercises and records
This part includes
- Metric layers
- Read signals across their maturity windows
- Source roles
1 self-check question
An account is free; sign in to read this part.
Checking your membership…
No checkpoint finished in this lesson yet. Start with “Go to the exercise”.
Self-check not answered
FAQ
- How is this data analysis lesson different from a GA4 reporting lesson?
A GA4 reporting lesson teaches events, reports, and explorations. This lesson answers the weekly operating question: one metric got worse, can the number be trusted, does the issue look like traffic, page, product, profit, fulfillment, or retention, and what single next-week action can be reviewed?
- If add_to_cart drops sharply on one SKU, what comes first?
Do not change the button or cut price first. Check whether traffic is stable, whether the drop is mostly mobile, whether hero image or price changed, whether stock and shipping promises changed, and whether the GA4 add_to_cart event still fires. The example turns "conversion is down" into trust, page, product, and tracking checks.