Data Quality Management
Dimensions (know the boundaries)
The DMBOK core dimensions are completeness, uniqueness, timeliness, validity, accuracy, and consistency (some frameworks add conformity, precision, accessibility, believability).
- Accuracy - data correctly describes the real-world object.
- Validity - data conforms to a defined format, type, or domain/range rule.
- Consistency - no contradictions across stores, systems, or time.
- Completeness - no missing values or missing records.
- Uniqueness - no unintended duplicates.
- Timeliness - available when needed and current enough for the purpose.
The management cycle
Define quality requirements and policy → profile/assess → measure and monitor (scorecards, thresholds, DQ SLAs) → analyze root cause → improve (preferably correct at source) → govern and re-measure continuously. Data profiling is measurement; cleansing is remediation. They are not the same step.
Traps
- Accuracy vs validity: a well-formatted but wrong phone number is valid but inaccurate.
- Believing cleansing solves a quality problem; without root-cause process fixes, defects return.
- Measuring quality without agreed business requirements - you cannot call data "poor" without criteria.
- Mixing completeness (missing values) with referential integrity (orphan records) or uniqueness (duplicates).
- Forgetting that quality is defined by fitness for use in a specific business context.
Worked example
A customer table has 5% blank email addresses → completeness defect; measure = % populated; fix = capture-process change at source. Two percent duplicate customer IDs → uniqueness. Birth dates of 2145 → validity (range rule), unless compared to the person's true birth date, which would be accuracy. Same customer address differing between CRM and billing → consistency.