An ecommerce dashboard should help you decide what to do. If it contains dozens of numbers but no one can explain what changed or what to investigate, it is mostly decoration.
Data analytics is the skill of defining a question, finding suitable evidence, checking its reliability, and turning the result into a decision. Start with a question such as “Why did orders decline?” rather than “What charts can we build?”
Agree on the definitions first
Different tools can use different definitions of sales, orders, customers, and sessions. Write a short metric dictionary before combining their reports.
| Metric | A practical definition for this exercise | Important boundary |
|---|---|---|
| Session purchase rate | Sessions with at least one purchase ÷ eligible sessions | Use one consistent session definition |
| Average order value | Defined order revenue ÷ included orders | Specify discounts, refunds, shipping, and tax treatment |
| Customer acquisition cost | Defined acquisition spending ÷ new customers acquired | State which costs and period are included |
| Return on ad spend | Attributed revenue ÷ advertising spend | Attribution is not proof of incremental revenue |
| Repeat purchase rate | Customers with another purchase within a defined window ÷ eligible customer cohort | Give everyone enough observation time |
These are working definitions. Label your dashboard so readers understand exactly what it measures. Do not compare a new-customer acquisition metric with a platform cost-per-purchase metric as though they count the same outcome.
Establish where each number comes from
Use the order system to reconcile orders and refunds. Use web analytics to investigate behavior on the site. Use ad-platform reports to understand their delivery and attribution. Use accounting records when you need the financial view.
These systems will not always match. Consent choices, blockers, attribution windows, reporting delays, time zones, and event implementations can create differences. Investigate the difference before assuming one dashboard is the complete truth.
For example, an ad platform may credit an order to an earlier ad interaction while the web analytics report assigns it differently. Adding the attributed revenue from several platforms can count the same order more than once.
Verify the events that describe the journey
For GA4, common ecommerce events include view_item, add_to_cart, begin_checkout, purchase, and refund. Google’s ecommerce measurement documentation describes the associated product and transaction data. A basic page-view setup does not by itself demonstrate that all of these events are implemented correctly.
Trace a supported test order. Check the item identifier, quantity, currency, and value. Confirm that the purchase uses a stable transaction identifier and that retries or page reloads do not inflate reporting. Verify refunds against the original transaction. Follow the integration’s documented behavior instead of assuming a theme or plugin handles every case.
Keep private customer information out of event names, URLs, and ad-hoc analysis exports. You rarely need an email address to determine where a checkout funnel is losing people.
Diagnose a decline with a worked example
Suppose a fictional store had 10,000 sessions and 200 purchasing sessions in one period, then 12,000 sessions and 180 purchasing sessions in the next. Its session purchase rate moved from 2% to 1.5%: a decline of 0.5 percentage points, or 25% relative to the earlier rate.
That does not explain why it happened. Break the comparison down by device, landing page, source, and product availability. Perhaps the store bought more exploratory traffic. Perhaps mobile checkout broke. Perhaps its bestselling variant sold out.
Check tracking changes before diagnosing customer behavior. Then form a small number of explanations and collect evidence that could distinguish them. Avoid slicing a tiny dataset into dozens of groups and treating every difference as meaningful.
Add an economic view
A sale can increase revenue while contributing little toward running the business. Alongside revenue, examine product cost, payment fees, fulfillment costs, shipping subsidies, discounts, and expected returns using definitions agreed with whoever owns your finances.
For a simplified example, a $60 order with $36 in variable costs leaves $24 before advertising and fixed costs. Spending $30 to acquire that order produces a negative first-order contribution under those assumptions. Possible repeat purchases do not repair the calculation until there is credible evidence and an acceptable payback plan.
This is a planning example, not a universal margin benchmark. Your actual costs and revenue recognition may differ.
Run a weekly decision review
Ask four questions: what changed, is the data credible, which explanations fit, and what action would teach us most? Assign an owner and a review date. Keep a note of the answer rather than letting the same unexplained issue return every week.
Use AI for a structured first pass:
Review this anonymized aggregate table and metric dictionary: [inputs]. Identify material changes, possible data-quality problems, and three competing explanations. Show calculations. State which additional evidence would distinguish the explanations. Do not claim causation or forecast revenue from this table alone.
Your assignment: define five metrics, validate one purchase journey, and write one decision brief. Then use the CRO guide to investigate friction or the advertising guide to connect acquisition costs with order economics.