10 Data Quality Checks to Run Before You Trust a Report

A polished dashboard can still be wrong. Clean charts and tidy columns tell you how information is presented. They do not tell you whether the underlying records are complete, current, or logically sound.

That distinction matters whenever a report influences staffing, resident follow-up, employee priorities, spending, or growth plans. Before you trust the conclusion, check the evidence underneath it.

Why data problems stay hidden

Most data errors do not stop a report from loading. A missing response date does not crash a spreadsheet. A duplicated resident does not trigger an alarm. An old staffing export can still produce a convincing trend line.

The report usually works. It simply answers the question incorrectly.

These ten checks catch many of the problems that distort day-to-day analysis. You can run them in a spreadsheet, a database, or a reporting tool. The tool matters less than the habit.

The 10 checks

1. Completeness

Look for blank values in fields that the analysis depends on. Start with identifiers, dates, outcomes, property or department labels, and the measures used in the final calculation.

A missing optional comment may be harmless. A missing renewal response in a renewal-intent report is not.

Ask:

• Which fields are required for this decision?
• What percentage of records is missing each required field?
• Are missing values concentrated in one property, team, period, or collection method?

2. Uniqueness

Confirm that records expected to be unique actually are. Common examples include survey response IDs, employee IDs, lease IDs, work orders, invoices, and customer transactions.

Duplicates can inflate counts and make a problem look larger than it is. They can also make one person's response carry more weight than everyone else's.

3. Accepted values

Fields with a defined set of choices should contain only those choices. If a status can be open, in progress, or resolved, values such as done, complete, and closed out need to be standardized before reporting.

This check is especially important when several people enter data or when two systems use different labels for the same idea.

4. Relationships

Connected records should agree. Every survey response assigned to a property should match a valid property. Every maintenance ticket should connect to an existing request. Every staffing record should map to a real role or department.

Broken relationships often reveal a process issue, not just a technical one. Find out why the connection failed before repairing the report.

5. Business rules

Test whether each record makes sense in the real world. A resolution date should not come before the issue date. Occupied units should not exceed total units. Hours worked should not be negative. A former employee should not appear in a current headcount unless the reporting rule explicitly includes them.

Write these rules down. A documented rule can be tested consistently. An assumption in someone's head cannot.

6. Reasonable ranges

Review minimums, maximums, and unusual values. A satisfaction score outside the survey scale is invalid. A response time of 900 days may be a unit error. A sudden workload value ten times larger than normal deserves inspection.

Do not automatically delete every outlier. Some extreme values are real and important. Investigate first, then decide whether to correct, exclude, or explain them.

7. Data types and formats

Numbers should be stored as numbers, dates as dates, and categories as consistent text values. Mixed formats create quiet failures in sorting, filtering, and calculations.

Pay attention to:

• Currency symbols stored inside numeric fields
• Dates entered in more than one format
• Percentages mixed with whole numbers
• Leading or trailing spaces in category names
• IDs that lose leading zeroes

8. Freshness

Record when each source was last updated and compare that time with the reporting period. A live-looking dashboard built on last week's export is still last week's information.

Freshness requirements should match the decision. A yearly workforce plan may not need hourly updates. A queue of unresolved resident concerns probably does.

9. Time consistency

Make sure systems use the same time zone, reporting window, and definition of a period. "Last 90 days" might mean a rolling window in one report and the previous three calendar months in another.

Small timing differences can create large disagreements near month-end, quarter-end, or a lease-renewal cutoff.

10. Reconciliation and sudden change

Compare totals with a trusted source and investigate sharp changes before explaining them. If completed surveys drop by half, check collection activity and system loads before concluding that residents stopped responding.

Useful comparisons include:

• Source-system totals against report totals
• Current period against previous periods
• Property or department totals against the portfolio total
• Raw response counts against scored response counts

Turn the checklist into a routine

The strongest data-quality process is repeatable. Assign an owner, document each rule, record failures, and decide what should happen when a check fails.

Use three levels:

1. Block: The report should not publish because a critical field or total is wrong.
2. Warn: The report can publish, but the limitation must be visible.
3. Monitor: The issue is not harmful yet, but a trend could become a problem.

This keeps quality work proportional. Not every imperfect value deserves an emergency, but every important limitation deserves an honest explanation.