QUALITY OPERATIONS / PRACTICAL FIELD NOTE

Write data quality rules your teams can actually test.

Replace generic accuracy percentages with clear populations, tolerances, owners and exception responses.

An engraved compass, reviewed records and calibrated scales; a metaphor for explainable data quality.
CONCEPTUAL ENGRAVING / DATAGOVERNANCE.CA

Practical information and fictional examples for scoping, not claims of client work, measured results, independent assurance or certification.

Name the business consequence

A “quality problem” matters because it changes a decision, causes repeated work or creates an avoidable exposure. Describe the exact workflow and consequence. A duplicate in a marketing export may mean something different from a duplicate in invoicing. Do not assume that a single company-wide accuracy score would help either owner.

Define the population and rule

A usable rule states the asset and field, unit of count, reporting period, exclusions, expected condition and tolerance. Example: in an agreed monthly billing population, each approved invoice should have one unique invoice identifier. The percentage of records that meet that condition is meaningful only with a stated denominator and treatment of missing and excluded records.

Decide what a failure means

Some failed checks require immediate containment; others trigger review before the next report. Specify who investigates, who can accept an exception, what supporting evidence is needed, and whether a report may still be used with disclosure. A threshold should be derived from business consequence and agreed tolerance, not selected because 99% looks impressive.

Test and re-evaluate

Use authorised sample records and deliberately include edge cases: late arrivals, backdated corrections, merged accounts, empty values and source outages. Record test results and dependencies. Re-run after relevant source changes and document any approved rule-version changes. The rule is only operational when someone follows up on failures and knows when to escalate.

QUESTIONS BEFORE DECIDING

What buyers usually ask

Should the target always be 100%?

No. Tolerance should match the business purpose, with explicit exceptions and their consequences.

Does a passed rule certify the report?

No. It supports a scoped assertion about a defined population.

NEXT PRACTICAL STEP

Start with one decision, not a platform purchase.

We can discuss a scoped engagement using the context above. Do not send confidential information through an initial inquiry.

READ NEXT