Data Quality Management
What this domain tests
Can you define, measure and remediate quality with business owners rather than run a one-off cleanse? Expect questions on dimensions, thresholds, root cause and issue management.
Core rules
- Quality is fitness for purpose, defined against a use. The same field can be fit for marketing and unfit for billing.
- Dimensions: completeness (populated), validity (conforms to domain or format), accuracy (matches reality), consistency (agrees across systems), uniqueness (no duplicates), timeliness (available when needed), conformity (standard formats).
- Lifecycle: profile → define rules with thresholds → measure and score → root-cause → remediate at source where possible → monitor continuously.
- Validity is not accuracy: a syntactically valid email can still be wrong. Only reference data or confirmation proves accuracy.
- Each rule needs an owner, a threshold, a frequency and an action, for example email completeness of at least 98% measured weekly, with a steward-opened issue below threshold.
- Prefer source correction over downstream cleansing; cleansing hides the defect and it recurs on the next load. Preserve the raw value for audit.
- Issue management: log, classify severity, root-cause (process, entry, integration, definition), fix, verify, close.
Worked example
A CRM email field is 82% complete; the campaign team needs 95%. Reasoning: (1) Define completeness precisely - non-null, valid format, not a role address - because counting non-null alone inflates the score. (2) Measure by segment; if B2B accounts drive the gap, target intake. (3) Root cause: email is optional at point of sale and unvalidated. (4) Fix at source: require it for B2B, add format validation, backfill through a consent-compliant campaign. (5) Set a weekly scorecard threshold and re-measure. Note the trade-off: forcing completeness can produce fake entries unless paired with validation.
Traps that cost marks
- Reporting a single data quality score with no dimension or denominator.
- Measuring only in the warehouse, after the defect has propagated.
- Confusing accuracy with validity or completeness.
- Cleansing without root cause, so the defect returns next cycle.
- Thresholds with no owner or action; a red dashboard nobody acts on.