Choose a decision before choosing a dashboard. Identify the source records, agree on metric definitions and compare consistent periods. Give the resulting action an owner and a review date; keep missing data, estimates and recorded results visibly distinct.
Business intelligence turns business records into decisions people can act on. It connects sources, defines the measures and shows what changed. You need it when separate reports leave the team arguing about what a number means.
1. Start with a business question
Choose a decision with an owner: which products need attention, where checkout stalls or whether a campaign is producing useful demand. Write what could change after the answer.
2. Identify the source records
Name the system behind each measure. Store orders, search impressions and platform-attributed purchases represent different records. Preserve that distinction when connecting them.
Google's GA4 Data API guide describes reports built from a specific property, dimensions, metrics and date ranges. A connection returns data; it does not decide whether the comparison makes business sense.
3. Define the measures and dates
Write what each metric includes, excludes and counts. Specify the time zone, period and source freshness. Distinguish recorded results from projections. Show incomplete coverage rather than treating missing records as zero.
For purchase measurement, check the GA4 purchase specification and trace an order using our event validation guide. Keep those definitions available alongside the report.

4. Build the view around the decision
Put the comparison, relevant breakdown and source date where the operator can see them. Identify attribution limits and incomplete records in the view itself.
In client work, our team built reporting into a staff application alongside assignments and service-history records. The public website was a separate part of the engagement. A prospective buyer needs an explanation of the offer; an operator needs records that support the next task. Decide which person the report serves before deciding what belongs on its first screen.
For a procurement-focused example, see ZINC Fusion v16, a soybean-oil forecasting platform in development. Its project overview describes market dashboards, purchasing scenarios and policy and sentiment views, with source dates, forecast horizons and data-availability context.
5. Assign an action and verify it
Record the observation, proposed action, owner and review date. After the change, compare the same defined measures over a suitable period. A report should help test a decision, not claim the decision worked before evidence arrives.
Decision record: question; source; metric definition; coverage dates; finding; limitation; action; owner; next check.
Start with one question through business intelligence. The channel planning guide helps choose which marketing decisions deserve attention first.
Define one measure before drawing the chart
Illustrative question: did the store’s net sales improve? Agree whether net sales subtract discounts and refunds, whether tax and shipping are excluded, which currency applies and whether refunds belong to their processing date or the original order cohort. Different valid definitions answer different questions.
Here is an illustrative definition and sample: net merchandise sales for an order cohort = merchandise charged after discounts minus merchandise refunds for those same orders as of the review date. Exclude tax and shipping from both sides. Use USD, the store's CT order dates and a stated refund-review cutoff. This is a cohort definition; it differs from subtracting all refunds processed today from today's new orders.
| Order | Merchandise after discounts | Merchandise refunded by cutoff | Net merchandise sales |
|---|---|---|---|
| A | $80 | $0 | $80 |
| B | $100 | $20 | $80 |
| C | $60 | $60 | $0 |
| Total | $240 | $80 | $160 |
A dashboard showing $240 has not yet applied these refunds. A dashboard showing $150 might include an unrelated $10 refund from a different order cohort. Neither discrepancy is solved by choosing the more flattering total. Trace the contributing records and use the agreed definition. Keep later refunds as dated revisions instead of pretending the earlier observation never existed.
Take a small sample of source orders and calculate the agreed measure by hand before trusting the dashboard total. Record source, grain, formula, exclusions, time zone, freshness and owner. If an order feed is missing, show the incomplete coverage rather than a zero. The acceptance test is that a second person can reproduce the result and understand what action it supports.
Make the definition reproducible
Use one row per measure. Record the source, grain, formula, exclusions, time zone and cutoff, then reconcile a small sample independently. The specimen uses the article’s fictional $160 order-cohort example; it is not a live sales report.
Download the metric-definition worksheet. Open the CSV in your spreadsheet application, keep the specimen labelled, and add your own rows. Pending observations are work to complete, not failed or successful results.
Sources
- GA4 Data API guidedevelopers.google.com
- GA4 purchase specificationdevelopers.google.com
