Data Governance
Data Governance
What is data governance?
- Managing data to be accurate, secure, and consistent
- A framework of rules, roles, and procedures
- The trust foundation for analytics and AI
What data governance really changes is not the system but where questions land. Before, the question 'why does this month’s customer count differ from report to report' all went to the IT team, and IT could only answer, 'We pulled it to the criteria the business gave us.' Once governance is in place, the authority to define and change that metric is explicitly assigned to the head of the business department. Knowing exactly who to call when a number is wrong is the first change.
In practice, roles fall into three layers. The data owner is the business executive for that domain, accountable for definitions and access approval; the data steward is a practitioner under them who regularly checks data quality; and IT handles technical custody such as storage, backup, and change history. Nor does the scope start company-wide. Even with thousands of tables, it is more realistic to begin with the 30–50 key metrics that reach executive meetings and master data (customers, products, vendors).
It is easy to confuse with neighboring concepts. Information security aims to 'keep things out,' but governance also covers opening data up properly to the people who need it. MDM and data catalogs are just tools that make the rules work; buying a tool does not give you governance. Quality is measured in numbers, not gut feel. Without metrics tracked monthly, such as duplicate-record rate, missing-value rate for required fields, and number of definition conflicts, governance stays a declaration rather than a practice.
The most common mistake in practice is treating the day the rulebook is rolled out as the finish line. You can circulate a 120-page standards document, but if the Excel templates the business teams use and the ERP code values stay the same, the numbers drift apart again within three months. The opposite failure is just as common. After one security incident, all permissions get locked down, and the marketing team needs four levels of approval just to view customer data. People end up going back to last month’s Excel file saved on a personal PC. Control hasn’t gotten tighter; the data has simply moved outside anyone’s management.
Ara LimExpert in global leadership, AI, and business communication
John KimAuthor of B2B Marketing Design and CEO of Nanum ConsultingIt grew out of the field of data management and became a core challenge as big data, AI, and privacy regulation spread.
Consider a hypothetical case. Company A makes industrial components, with 900 employees and annual revenue of KRW 320 billion (≈ US$239,413,400). ERP, MES, and CRM each ran on their own, and customers were registered separately in each system. In the monthly sales meeting, the list of top customers by revenue differed from department to department, and the same argument had been repeating for six months.
- When the head of Management Support cross-checked customer codes across the three systems, roughly 800 actual companies turned up as 1,450 records. In the meeting someone asked, 'The same company is registered three times, so how can we trust the sales ranking?'
- They named the head of Sales as the owner of customer data and assigned a sales support manager as steward. From March, no new customer record or change could be processed without both people’s approval.
- Sales had always called a company 'a customer' once a quote went out, while Accounting used the term for companies that had been issued a tax invoice. They split the two definitions into 'prospect' and 'active customer' and started by renaming the dashboard fields.
- Access moved to a tiered model instead of a full lockdown. Assigned clients can be viewed instantly, the full list needs one-step manager approval, and access logs are pulled each quarter — about 40 accounts unused for six months were closed.
- They set duplicate rate and missing-value rate for required fields as monthly metrics, with the head of IT reporting on the first business day of each month. Any item over the threshold had to be cleaned up by the steward within two weeks.
- Eight months later, duplicate client records fell from 1,450 to 820, and the time to finalize sales results after month-end close dropped from 5 days to 2.