Who is responsible for data quality in your company?
If no one is responsible for the quality of a piece of data, that data is not reliable by definition.
Data governance is not bureaucracy. It is the answer to four questions that almost always remain undocumented in a company: what data do we have and where is it stored, who is responsible for each piece of data, who can access it, and how long do we keep it.
The benefit is twofold: decisions are based on numbers you can trust, and when an audit or a customer request about their data comes in, you can respond in hours rather than weeks.
A catalog and a glossary, not the memory of whoever has been here the longest
The catalog shows where each piece of data lives; the glossary defines what it means within the company. It seems redundant until you discover that “active customer” has three different definitions across sales, administration, and marketing — and that the three conflicting reports were all correct according to their respective definitions.
The catalog is populated automatically from the data pipelines, not through manual entry: a catalog maintained by hand becomes outdated within a quarter.
Who is responsible for what
- Data owner: defines the meaning, criticality, and access rules for the data. This is a business role, not a technical one.
- Data steward: ensures the rules are applied and oversees data quality on a day-to-day basis.
Without these two roles, quality becomes everyone's responsibility, and therefore no one's. We define them together with you during the initial phase, focusing on the data domains that actually matter.
Quality measured, not declared
Automated checks run on every data load, with the results recorded: required fields populated, valid formats, keys present in reference tables, values within plausible ranges, and no duplicates. A failed check triggers an alert and, in critical cases, blocks downstream publication rather than allowing the invalid data to spread.
The measurements are stored over time because the absolute value matters less than the trend: a gradual decline is a signal that something has changed upstream.
Access and personal data
Permissions follow business roles, not custom: salespeople see their customers, managers see their area, and management sees the aggregate. When someone joins, they get access to what they need for their work; when they leave, they lose it.
Personal data that is not required for analysis is masked at the source, rather than made available and then ignored by agreement. The rules apply within the data platform and therefore to every tool that accesses it, including natural language queries.
Retention and traceability
For each data category, we define the legal basis for processing, the retention period, and what happens when it expires. Automatic deletion at the end of the retention period is preferable to deletion on request, which no one remembers to carry out.
Tracing the journey of every piece of data — where it comes from, what transformations it has undergone, and where it ended up — serves three concrete purposes: assessing the impact of a change before making it, identifying the cause of an incorrect figure, and demonstrating data provenance during an audit.