Data Quality Management
Data Quality Management
What is data quality management?
- Block errors at entry instead of fixing them later
- Put an unchanging identifier, such as employee ID, in the first column
- The department that uses a metric reviews it monthly
What data quality management changes is that the question 'Is this number right?' disappears from meetings. When headcount in the HR system differs from headcount in payroll, training completion lists include people who have left, and the same customer is registered under three names, no analysis can be trusted. Quality management turns these mismatches into rules that stop errors at the point of entry, rather than fixing them after the fact. No matter how good AI and BI tools become, wrong inputs produce wrong results, which is why quality is the precondition for every use of data.
Quality is usually judged on six dimensions: accuracy (does it match the real value), completeness (no blanks), consistency (the same across systems), timeliness (updated on time), uniqueness (no duplicates), and validity (fits the defined format). In practice, each dimension gets check rules and acceptance thresholds expressed as numbers: email format errors at or below 1%, zero missing department codes, weekly updates, and so on. If data governance is the framework that defines 'who is accountable,' quality management is the day-to-day work of measuring and fixing data within it. Internationally, DAMA's DMBOK codifies this area.
Training and HR data are especially prone to quality problems. Training records are scattered across the LMS, Excel files, and certificates from external providers, and reconciling statutory training completion all at once at year-end leads to omissions. During evaluation season, evaluator-to-employee mappings fail to keep up with organizational changes, so the wrong department head ends up rating someone. Merging data by name mixes up people who share a name, so putting an unchanging identifier such as an employee ID in the first column of every table is half of quality management.
The most common failure in practice is treating quality management as a one-time cleanup. You can remove duplicates and fill blanks at year-end, but if the input rules stay the same, the data is back in the same state three months later. Another frequent mistake is making data 'clean' by overwriting the original, so past values can never be recovered. If you hold the business teams who enter the data accountable for quality without explaining why it matters, you get perfunctory entries, such as 'N/A' or '-' in required fields. Quality metrics stick only when they land on the monthly agenda of the departments that use the data, not just IT.
Jaryong KooMarketing PhD and authority in AI, data analytics, and personal branding
John KimAuthor of B2B Marketing Design and CEO of Nanum ConsultingQuality management became its own discipline in the 1990s, when databases moved to the center of business operations and the cost of data errors to decision-making drew attention. The DMBOK (Data Management Body of Knowledge), published by the professional association DAMA, is the standard reference that defines quality dimensions and management procedures.
Consider a hypothetical case. Company A, a pharmaceutical distributor with 730 employees, spent three sleepless nights last December merging data from its LMS, Excel files, and external certificates to submit its statutory training completion status for a regulatory inspection. The table it submitted listed 14 former employees as non-completers and left out 6 current employees. The HR director vowed not to go through that again this year.
- When Ms. Seo in HR matched the lists from the three sources by employee ID, she found that merging by name had mixed up 9 pairs of people with the same name, and the external certificate file had no employee ID column at all.
- They set a rule that the employee ID is the first column of every training list, and added an employee ID field to the enrollment forms sent to external providers so the ID comes back attached to each certificate.
- They set five quality rules with numeric targets: 0 missing employee IDs, 0 former employees included, department-code mismatches of 1% or less, 0 missing completion dates, and a monthly refresh. The IT team built a checklist that runs these rules automatically every week.
- The first check found an 11% department-code mismatch. Ms. Seo, the HR associate, traced it to a simple cause: nobody had updated the LMS department codes after the reorganization. She added a monthly step to match HR transfer records against the LMS org chart.
- This December, the statutory training completion report was ready in half a day, with 0 former employees wrongly included and 0 current employees missing. Evaluator-mapping errors also fell from 23 last year to 2.