A polished dashboard can still be wrong. Missing transactions, duplicate customer records, stale refreshes and inconsistent definitions may produce confident-looking numbers that cannot support a decision. Data quality assurance is therefore part of an analyst’s job, not a separate activity performed only by engineers.
This checklist focuses on the tests an analyst can apply before a dashboard reaches stakeholders.
Confirm the business grain
The grain describes what one row represents: one order, one order line, one customer per day or one account per month. Write it down before joining tables. If an order-level table is joined to an order-line table, revenue may be repeated for every item unless the aggregation is designed correctly.
Test expected uniqueness at the stated grain. A primary key count should equal the distinct-key count. If it does not, investigate before building measures.
Reconcile important totals
Compare dashboard totals with a trusted operational or finance source for the same period and filters. Reconcile record counts, revenue, refunds and active customers separately. A matching grand total does not prove that categories or dates are correct.
Document tolerances. Currency rounding may create a small permitted difference, while a customer-count difference may require exact agreement.
Test completeness
Completeness checks ask whether expected data is present. Review null rates in key fields, missing dates, absent regions and files that failed to arrive. A sudden drop in daily records may be a pipeline issue rather than a real business decline.
Create a calendar comparison showing expected dates against loaded dates. For scheduled feeds, log the latest successful ingestion time and expected delivery frequency.
Check validity and consistency
Validity means values follow permitted rules. Examples include positive quantities, recognised status codes, valid dates and email formats. Consistency means the same concept is represented the same way across sources.
Visakhapatnam, Vizag and VISAKHAPATNAM may refer to one location but create three categories. Create controlled mappings rather than silently correcting labels inside individual charts.
Find duplicate records
Duplicates can enter through repeated file loads, one-to-many joins or weak source keys. Search for both exact duplicates and business duplicates. Two rows may have different technical IDs but the same customer, timestamp and transaction amount.
Do not delete duplicates automatically. First determine whether they represent repeated data, split transactions or legitimate events.
Verify freshness
Display a “data updated at†timestamp and compare it with the service expectation. A daily dashboard refreshed three days ago should not be interpreted as today’s performance.
Freshness also depends on completeness. A pipeline may finish successfully while loading only part of a file. Track maximum event date, row counts and control totals together.
Validate calculations and filters
For every major KPI, independently reproduce a small example. Test date boundaries, null handling, returns, cancellations and divide-by-zero cases. Confirm that slicers affect all intended visuals and do not unintentionally change fixed benchmarks.
Check whether percentages use the correct denominator after filtering. A regional share may need total company revenue, not merely the filtered region total.
Review outliers with context
Outliers are not automatically errors. A large enterprise order can be real. Flag unusual values and ask for business confirmation before excluding them. Keep an audit trail of exclusions and show how they affect the result.
Create a release checklist
Before publishing, record the dataset version, refresh time, row count, reconciliation result, unresolved issues, report owner and approval. Automate repeatable tests in SQL or Python, but keep human review for business definitions and unusual events.
The Data Analytics Course in Vizag covers data cleaning, SQL, Python and dashboard workflows that support these checks. Apply the checklist to a retention cohort analysis and verify any experiment dashboard with the A/B testing guide for data analysts.
Final takeaway
Reliable dashboards are built on explicit grain, reconciled totals, complete records, valid categories and visible freshness. A short quality log can prevent hours of debate and protect stakeholder trust. Treat every surprising result as a finding to verify before it becomes a recommendation.