Seven Data Quality Checks Every Dashboard Should Run Before You See It

A dashboard is only as good as the data underneath it, and bad data rarely looks bad. A chart built on stale or incomplete numbers looks exactly as polished as one built on correct numbers. The difference is that people make decisions from it.

The fix is to check the data before anyone sees it. Here are seven checks worth running on every dashboard.

Freshness

Is the data as recent as it should be? Record when each source last loaded and show it on the dashboard. A data as of time costs almost nothing and stops people acting on yesterday's figures without realizing it. Set a threshold, and when it is exceeded, warn people clearly instead of silently showing old numbers.

Completeness

Did all the expected rows arrive? Compare row counts with what you normally see for the period. A sales table that has half its usual rows on a Monday morning usually means a failed load, not a quiet weekend.

Duplicates

Do any records appear twice? Duplicates inflate totals and are common after a retried load or a join that multiplied rows. Check that the fields that should be unique, such as an invoice number, actually are.

Missing values in key fields

Which important fields are blank? A sales order with no customer or a transaction with no date will fall out of some reports and not others. Count blanks in the fields your numbers depend on.

Orphaned records

Do records point to things that do not exist? An order line for a product that is not in the product table will drop out of any report grouped by product. Check that references line up.

Sudden changes

Is anything far above or below normal? A day with ten times the usual revenue or none at all is either extraordinary or an error, and it is worth a person looking before it goes into a board pack. Compare each day with recent history and flag the outliers.

Reconciliation to the source

Do the totals match the system of record? Pick a few numbers, such as monthly revenue or inventory value, and compare the dashboard total with the figure in the ERP or the ledger. If they tie out, trust grows. If they do not, you find the reason before your readers do.

Make the results visible

Checks that nobody sees do not help. Show a simple status on the dashboard: passed, passed with warnings, or failed. When a check fails, tell the person who owns the data, not only the person reading the chart. On a plant floor, where operators notice a wrong number immediately, this matters even more, as we describe in how to build a production dashboard your operators will actually use.

Horizon runs checks like these hourly on each report's headline numbers and shows whether the latest data passed, so a person can see at a glance whether to trust what is in front of them.

Start with three

You do not need all seven on day one. Freshness, completeness and reconciliation to one trusted total catch most of the problems that actually embarrass people. Add the rest as you learn where your data breaks.

If your team has been burned by a dashboard that turned out to be wrong, reach out. Building the checks in from the start is much cheaper than rebuilding trust afterwards.