“Good governance is not about control. It’s about creating the conditions where trust, accountability, and value emerge naturally from the way people already work.” — Bob Seiner, Non-Invasive Data Governance
From Chaos to Clarity: Evolving Data Governance Without Killing Innovation
It always starts the same way. A fast-growing company realizes that its analysts are spending more time arguing about which numbers to trust than making decisions. Marketing has their “customer count.” Finance has theirs. Product swears the data in their dashboards is right—except no one can explain the SQL buried three layers deep in Looker.
This is the moment most organizations realize they don’t just need more data. They need better governance.
A Brief History of Data Governance
Data governance has historically carried a bad reputation. In the 1990s and 2000s, it was synonymous with bureaucracy: councils, policies, and binders that slowed projects to a crawl. Entire committees debated who could access a table instead of asking how to help people use it effectively.
The backlash was predictable. Business units rebelled, shadow IT flourished, and self-service BI tools like Tableau and PowerBI exploded in popularity. But with that freedom came new chaos: duplicate data, inconsistent definitions, and regulatory risks.
Enter a new generation of governance thought leaders—John Ladley, Gwen Thomas, Bob Seiner—who reframed governance not as a police force, but as an enabler. Seiner’s “Non-Invasive Data Governance” became a rallying cry: governance shouldn’t take away what people already have; it should give them something better.
Types of Data Governance: From Centralized to Federated
Centralized Governance
The traditional model: one team (usually IT or data management) owns definitions, policies, and approvals. It works well for highly regulated industries like banking but often creates bottlenecks.
Federated Governance
Emerging with the rise of data mesh (championed by Zhamak Dehghani), federated governance distributes responsibility to domain teams. Instead of one group dictating terms, each domain owns their data products—definitions, quality, and compliance—while adhering to common standards. Think of it as federal vs. state government.
Non-Invasive Governance
Bob Seiner’s approach: meet people where they are, document the governance they’re already practicing (even informally), and incrementally improve it. It’s adoption by stealth—rather than imposing a grand plan, you show people how small changes make their lives easier.
The most successful organizations blend these approaches. Centralized standards (naming conventions, access controls) combine with federated ownership (teams accountable for their own data products) and non-invasive rollout (giving teams better tools and reducing friction).
Where to Start: Zero to One
Most companies start at zero—no consistent definitions, no lineage, no clarity on who owns what. Here’s how to climb the ladder:
- Stage 0 – Chaos
- Data exists, but trust doesn’t. Metrics differ by department. Access is haphazard.
- Example of failure: A global retailer where analysts manually export spreadsheets every week from three different CRMs—producing three “official” customer counts.
- Stage 1 – Inventory and Ownership
- Create a data catalog or lightweight wiki of key data assets. Assign informal stewards in each department.
- What good looks like: A health-tech startup that adopted Atlan as a collaborative catalog, letting analysts add context (“this field comes from Salesforce, last updated daily”) before worrying about strict policies.
- Stage 2 – Definitions and Standards
- Establish core business metrics (customer, ARR, churn) with executive buy-in. Create simple access standards (PII handling, role-based access).
- Failure case: A financial services firm that launched a 200-page policy doc but never explained to teams why the new process was better than what they had—leading to wholesale rejection.
- Stage 3 – Automation and Quality
- Introduce data quality checks, lineage tracking, and monitoring (Monte Carlo, Soda, dbt tests). Policies are enforced in pipelines, not PDFs.
- Success: Netflix’s federated data quality framework, where domain teams define tests but infrastructure automates enforcement.
- Stage 4 – Federation and Self-Service
- Domain teams own their data products, aligned to global governance standards. Access, lineage, and compliance are discoverable in-platform.
- Good example: Airbnb’s Minerva project—a federated metric platform ensuring consistent definitions across teams while keeping ownership decentralized.
- Stage 5 – Embedded Governance
- Governance becomes invisible. People don’t think “I need to comply with governance”; they just see trusted, high-quality data when they need it.
Milestones That Drive Maturity
- Regulatory pressure: GDPR, HIPAA, CCPA often force the jump from Stage 1 → Stage 3.
- Organizational scale: Once you pass 50 analysts or multiple product lines, federation becomes inevitable.
- Technology shifts: Moving to a lakehouse or adopting a data mesh requires governance by design.
The Adoption Secret: Always Offer Something Better
The biggest mistake is rolling out governance as a restriction. People resist losing autonomy. Instead, the question should always be: What pain am I removing?
- Instead of taking away analysts’ access, give them a catalog that makes data easier to find.
- Instead of telling engineers to follow new PII policies, give them APIs that mask data automatically.
- Instead of mandating new metric definitions, provide pre-built, certified dashboards that save them hours.
Change adoption happens not when governance tells people what they can’t do, but when it gives them something better than the messy status quo.
Wrapping up…
Data governance isn’t a project. It’s an evolving capability. Organizations that succeed don’t treat it as a compliance exercise—they treat it as a way to enable trust, speed, and scale.
From chaos to clarity, from centralized control to federated ownership, and from restrictive bureaucracy to non-invasive enablement—the arc of modern governance bends toward empowerment.
And the real milestone? When no one talks about “data governance” anymore. They just trust the data.
