Which decision belongs on this page?
A dashboard starts earning its place when someone can use it to make a recurring decision. A marketing lead may need to identify campaigns producing qualified enquiries. An operations manager may need to find enquiries waiting too long for a response. Both can ask for a performance dashboard, but they need different records, comparisons and actions.
Write the reader, decision and frequency in one sentence before choosing charts. For example: each Monday, the marketing lead compares completed acquisition cohorts to identify channels that need further investigation. Then ask what the reader would do if the number increased, decreased or stopped updating. If the answer is unclear, the metric probably needs a better definition or a less prominent position.
For a new dashboard, this sentence limits scope. For an existing one, use it to review every visible panel. A chart may be accurate and attractive yet contribute nothing to the decision. Archive unused views so their logic remains available without making the main page harder to read.
What does each metric include?
Create a short metric contract. For qualified enquiries, specify the unit being counted, the qualifying statuses, who assigns them, the treatment of duplicates and the date that places an enquiry in a period. Add the source and the person responsible for changes. The label qualified leads cannot carry all of those definitions on its own.
Keep numerator and denominator together. A conversion rate might mean qualified enquiries divided by eligible sessions, qualified enquiries divided by submitted enquiries, or deals won divided by qualified enquiries. Each is useful for a different purpose. State whether the records form a cohort: deals closed this month divided by leads created this month can mix people who entered the funnel at different times.
Make exclusions visible in the definition. If staff tests, spam or specific countries are omitted, document the rule and where it runs. A filter added only inside one chart can produce two apparently identical metrics with different values.
Does the calculation survive a small example?
Suppose an illustrative campaign produced one enquiry from ten eligible sessions on Monday and nine enquiries from ninety sessions on Tuesday. Both daily rates are 10%, so their average and the combined rate agree. Now change Monday to one enquiry from ten sessions and Tuesday to one enquiry from ninety sessions. The daily rates are 10% and approximately 1.11%. Their simple average is about 5.56%, while the combined rate is two divided by one hundred: 2%.
Both calculations are mathematically valid, but they answer different questions. If the dashboard label means enquiries per eligible session across the selected period, use the combined numerator and denominator. Google's calculated-field documentation explains that fields can operate on individual rows or aggregated values. Check the formula at the intended level rather than relying on the number format or chart label.
Include empty and unusual inputs in the review. Decide what a zero denominator should display. Distinguish a true zero from missing data. Test whether changing the date range, dimension or filter changes the calculation as intended. A small hand-calculated dataset is often more revealing than an impressive-looking chart with thousands of rows.
Documentation: Google Cloud: calculated fields and aggregation
Can you trace the number through its sources?
List the source system, table or report, fields, refresh process and owner for each major metric. The reader does not need a database diagram on the first page, but should be able to reach a definition and supporting detail. The builder needs enough documentation to identify which integration or calculation to inspect when the number changes unexpectedly.
Joining sources needs a stated matching rule. Google's documentation on data blends explains that join conditions connect records and join operators control how matching and unmatched rows combine. Before choosing either, write what one row means in each input: one campaign per day, one lead, one order or one item are different levels of detail.
Imagine a table with one 100-unit spend record for a campaign and another table with three leads from that campaign. A raw join can repeat the spend beside each lead; summing those repeated values would produce 300. Aggregate both inputs to an agreed level or use a model that preserves the intended relationship. Reconcile totals before and after joining, and inspect unmatched records instead of silently discarding them.
Documentation: Google Cloud: how data blends work
How current is the underlying information?
Separate the time the viewer opened the dashboard, the last successful source update and the latest complete reporting period. A page loaded at 09:00 may be showing a source whose import stopped yesterday. A single updated now label can hide that distinction. Prefer a specific note such as CRM import completed at 08:15; yesterday's records included, when those facts are actually available.
Google's data-freshness documentation explains that the reporting tool can reuse stored query results within a source's freshness threshold and fetch again when appropriate. That cache behaviour is separate from whether the underlying source itself is complete. Changing a dashboard refresh setting will not repair an upstream import that has failed.
For period comparisons, display dates and identify incomplete days. Ask whether the periods contain comparable weekdays and whether the underlying definitions stayed the same. If a sales team changed its qualification rule, annotate the change. An unexplained jump after a definition change should not become a performance claim.
Documentation: Google Cloud: manage data freshness
What can the reader do with an unusual result?
Give a surprising number a route to investigation: a channel breakdown, the relevant enquiry queue or a note about a known source issue. Show the volume behind a rate. A change from one conversion to two and a change from one hundred to two hundred have the same percentage increase, but they call for different confidence and follow-up.
Use colour for a defined meaning. A red mark should indicate an agreed condition requiring attention, not every downward movement. Keep the first view focused on a few decisions, with detail available when needed. On small screens, preserve the reading order and labels; shrinking a wide desktop canvas until the figures fit rarely creates a usable mobile report.
Who will keep the dashboard useful?
Test the page with a realistic task and the intended reader. Ask them to explain a metric, identify what changed, find its source and choose the next step. Observe where they hesitate. Those pauses reveal missing definitions and navigation problems more directly than asking whether the design looks good.
Before handover, confirm that totals reconcile, filters behave consistently, empty states are readable, access is appropriate and failed refreshes have an owner. Store definitions and a brief change log beside the report. Schedule a review based on how often the business changes, with a clear route for reporting problems. A maintained dashboard should continue answering the decision in its brief after the person who built it has left the meeting.