Data quality is fitness for a stated purpose, and it can be measured. Pick the data elements that matter most, profile them, define rules that express what good looks like, measure against thresholds, and fix causes at the source rather than cleaning the same errors every month.
Six dimensions to measure
Standards such as ISO/IEC 25012 and ISO 8000 describe data quality through named characteristics. Six are enough for most programs.
| Dimension | Question it answers | Example measure |
|---|---|---|
| Completeness | Is required data present? | Records with a valid email divided by total records |
| Uniqueness | Is each real-world thing recorded once? | Duplicate records divided by total records |
| Validity | Does data follow the allowed format or range? | Dates in the expected format divided by all dates |
| Accuracy | Does data match the real world? | Sampled records confirmed against a trusted source |
| Consistency | Does the same fact agree across systems? | Customers with matching status in both systems |
| Timeliness | Is data current enough for its use? | Records updated within the agreed time window |
The quality cycle
Step by step
- Choose critical data elements. Start with 10 to 20 elements that drive revenue, risk, or regulatory reporting, such as customer identifier, contact details, and account status.
- Profile. Count nulls, distinct values, patterns, minimum and maximum, and frequent values. Profiling shows what the data really looks like before anyone argues about what it should look like.
- Define rules and thresholds. Turn requirements into testable statements, for example: email must match a valid pattern, and at least 98 percent of active customers must have one.
- Validate at entry and in pipelines. Block bad values at the form or interface, and run the same rules when data moves between systems.
- Score and report. Build a scorecard by data element and dimension. Show trend, not only a snapshot.
- Find root causes. For each failing rule, trace where the error enters: a form with no validation, a manual re-keying step, or a mapping defect.
- Fix, prevent, monitor. Correct the data, change the process that created the error, and keep the rule running.
A worked example
A customer table has 12,400 records. Profiling finds 11,780 with a valid email and 372 duplicate records. Completeness for email is 11,780 divided by 12,400, which is 95.0 percent. Duplicates are 372 divided by 12,400, which is 3.0 percent. The agreed thresholds are 98 percent and 1 percent, so both fail. Root cause analysis shows that the web form accepts any text in the email field and that two onboarding channels create customers separately. Adding format validation and a duplicate check at entry addresses both causes; a one-time clean-up handles the backlog.
Common pitfalls
- Measuring everything. Focus on the elements that matter to decisions.
- Cleaning without fixing the cause, so the same errors return.
- Rules written by technologists alone. Business stewards must define what good looks like.
- Reporting a single overall score that hides which dimension is failing.
Implementation checklist
- Critical data elements listed with owners
- Profiling completed on each element
- Rules and thresholds agreed with business stewards
- Validation active at entry points and in data pipelines
- Scorecard by element and dimension published monthly
- Root causes recorded and corrective actions tracked
- Alerts set for rules that fall below threshold
Treat data quality as a service level, not a clean-up project. Agree a threshold with the data owner, publish the result, and review misses the same way you would review a missed system uptime target.
Interactive: data quality calculator
Change the numbers to see how completeness and uniqueness are measured against agreed thresholds.
Email completeness
95.0%
Target: at least 98%
Below threshold
Duplicate rate
3.0%
Target: at most 1%
Below threshold
References and further reading
- ISO/IEC 25012:2008 Data quality model (ISO)Accessed November 7, 2026
- ISO 8000-61:2016 Data quality management: process reference model (ISO)Accessed November 7, 2026
- DAMA-DMBOK: Data Management Body of Knowledge (DAMA International)Accessed November 7, 2026