Minutes 0–5: read the contract
Find the dashboard owner, intended audience, refresh time, timezone, reporting currency, and definition of each headline metric. If those details do not exist, the dashboard is not decision-ready. Write them down before interpreting trends.
Confirm whether dates refer to event time, order time, lead creation, or reporting ingestion. Check whether the current day is partial and whether comparison periods contain the same number of complete days. Many apparent performance swings are simply mismatched windows.
Minutes 5–12: reconcile the outcome
Compare the main outcome—revenue, orders, qualified leads, or subscriptions—with the closest system of record. Use the same timezone, currency, refund rule, internal-traffic rule, and date window. The goal is not perfect equality; the goal is an explained difference.
Record both the absolute and percentage gap. Then sample individual records. A stable 4% gap caused by consent coverage means something different from a 4% gap caused by duplicated purchase events. Without a reason and an expected range, a tolerance is only a way to stop asking questions.
Trust is not “the totals match.” Trust is “we can explain why they do not match exactly.”
Minutes 12–18: look for duplication and loss
Compare unique transaction or lead identifiers with row counts. Repeated identifiers expose duplicate ingestion. Missing identifiers make deduplication impossible and should lower confidence immediately. Check a high-traffic page or funnel step for a sudden change in event-to-session ratio.
Inspect whether refresh failures create flat lines, whether late-arriving conversions rewrite prior dates, and whether bot or internal traffic is excluded consistently. A dashboard that silently changes historical totals needs a visible last-updated note and a documented restatement window.
Minutes 18–24: challenge channel classification
Look at the share of direct, unassigned, and unknown traffic. Review the largest source-medium pairs inside each channel. One broken UTM convention or changed platform identifier can move a meaningful share of demand without changing total demand at all.
Check whether paid clicks and analytics sessions move in the same general direction. They will not match one-to-one, but sharp divergence can expose consent changes, landing-page failures, redirects that strip parameters, or a campaign tagged into the wrong channel.
Minutes 24–30: test the decision
Take one decision the dashboard is meant to support and trace it from headline to source rows. Change a filter, inspect the denominator, and confirm that totals behave as expected. If the view cannot explain its own result, it should not be the only evidence used for action.
Finish with a trust note: what was checked, what reconciled, the known gaps, and when the test should run again. Show freshness and caveats in the dashboard itself. Data quality hidden in a separate document rarely reaches the person making the decision.