We talk a lot about data-driven decisions, dashboards and predictive models. But one truth often stays in the shadows: a decision is never better than the data it rests on. A beautiful dashboard fed by wrong data produces wrong conclusions, and all the more confidently because it looks polished. Data quality and governance are that quiet foundation, rarely in the spotlight, yet they decide whether the entire decision chain holds up or collapses in silence.
Data quality is a business issue, not a technical one
We tend to file data quality under "IT problem." That's a framing mistake. The consequences of faulty data are business consequences first: a marketing campaign sent to outdated addresses, a miscounted stock that triggers a shortage, distorted financial reporting that drives a bad investment decision. Gartner estimates the average cost of poor data quality at nearly $13 million per year per organization. And the most insidious cost isn't financial: it's the loss of trust. The day an executive spots one wrong figure in a report, they stop trusting every report, including the accurate ones.
The six dimensions of data quality
"Quality data" is a vague phrase. To make it actionable, it's classically broken down into six dimensions, each tied to a concrete question:
- Completeness: is all the expected data present? An empty "email" field for 30% of customers directly limits any campaign.
- Accuracy: does the data correctly reflect reality? An address that is valid in form but doesn't exist is an accuracy problem.
- Consistency: is the same information identical across systems? A customer "active" in the CRM and "closed" in billing creates a blocking inconsistency.
- Timeliness: is the data up to date when it's used? A stock level refreshed once a day is useless for a real-time decision.
- Uniqueness: does each entity exist only once? Duplicates distort counts and artificially inflate metrics.
- Validity: does the data follow the expected rules and formats? A birth date in 2043 is syntactically a date, but it's invalid.
The value of this breakdown is operational: it turns a fuzzy feeling ("our data is bad") into measurable rules you can track over time.
What is data governance?
If quality describes the state of the data, governance describes the organization that keeps it there over time. It's the set of roles, rules and processes that ensure data is reliable, understood and used correctly. Without governance, quality is a one-off cleanup that degrades the next day; with it, quality becomes a maintained state.
Governance usually relies on three complementary roles:
- The data owner: often a business leader, accountable for the data in their scope. They set the rules and arbitrate priorities.
- The data steward: the operational guardian of quality. They translate business needs into concrete rules, document the data, monitor its quality and bridge business and technical teams.
- The data custodian: on the technical side, they manage storage, access and security according to the defined rules.
The foundation's tools: catalog, glossary and lineage
Effective governance relies on a few key artifacts that make data understandable and traceable:
- The data catalog: a centralized inventory answering "what data do we have, where, and what does it mean?" Without a catalog, knowledge stays in a few people's heads.
- The business glossary: a shared definition of key terms. What exactly does "active customer" mean? If three teams have three definitions, their figures will never be comparable.
- Data lineage: tracing a data point's journey from source to dashboard. It reveals the impact of a change and lets you trace an anomaly back to its origin.
Where to start in practice
A governance effort can quickly feel overwhelming. The key is not to tackle everything at once, but to proceed in steps:
- Identify critical data: not all data carries the same weight. Start with the Critical Data Elements that feed your most important decisions and metrics.
- Profile what exists: measure actual quality before trying to improve it. Data profiling reveals completeness rates, duplicates and outliers.
- Define rules and automate them: translate each dimension into automated checks, run continuously rather than case by case.
- Build a culture: tooling isn't enough. Quality improves when every data producer understands they are the first person responsible for its reliability.
The link with compliance and GDPR
Governance isn't only about decision comfort: it's also a regulatory obligation. GDPR notably imposes a principle of accuracy (personal data must be accurate and kept up to date) and minimization (keep only the data you need). Knowing precisely what personal data you hold, where it lives and who accesses it (exactly what a catalog and lineage enable) is no longer an optional best practice; it's a legal requirement.
Conclusion
Data quality and governance are nothing spectacular. You can't run an impressive demo with a data catalog or a business glossary. Yet it's this groundwork that decides, upstream, the reliability of everything that comes after: the analyses, the dashboards, the models, the decisions. Treating governance as bureaucracy misses its true nature: it's what turns a mass of raw data into a trustworthy asset. And trust, when it comes to decisions, is exactly what carries the most value.