Published November 20, 2026
What data governance is, and why now
Data governance is the set of decisions about data and who makes them: who owns each data set, what its terms mean, what quality it must meet, who may use it, and who settles a dispute.
Data management is the work that carries those decisions out: pipelines, databases, quality checks, access controls.
Three pressures bring the job to the CIO. Analytics and AI projects need data that someone can vouch for. Privacy rules such as GDPR Article 30 require a record of what personal data is processed and why. Security frameworks ask for a data inventory: NIST CSF 2.0, published February 26, 2024, includes ID.AM-07, "Inventories of data and corresponding metadata for designated data types are maintained."
Retention schedules are on the data retention policy page. Governance of AI systems is on the AI governance page.
Where it sits, and who owns the data
The first decision is the operating model. A central team can set every standard and approve every change, which gives consistency and becomes a queue. Business domains can govern their own data, which scales and drifts apart. A federated model sets a small number of shared standards centrally and leaves ownership and day-to-day decisions in the domains.
The second decision is who leads: a chief data officer, where one exists, or the CIO. If it sits with the CIO, the business may read it as an IT project. IT can run the catalog and the controls. It cannot decide what "active customer" means for sales.
Name an executive sponsor outside IT, and a business owner for each data domain who can make decisions without escalating.
The frameworks
| Framework | Who publishes it | What it is |
|---|---|---|
| DAMA-DMBOK | DAMA International | A body of knowledge for data management, with data governance as one of its knowledge areas. The current edition is the 2nd, revised in 2024. A DMBOK 3.0 project started in 2025. |
| DCAM | EDM Council | The Data Management Capability Assessment Model: the scope of capabilities required to establish, enable and sustain a mature data management discipline. |
| CDMC | EDM Council | Cloud Data Management Capabilities: an assessment and certification framework for data in cloud environments. It includes 14 Key Controls and Automations for protecting sensitive data, and is free to license for internal use. |
| ISO/IEC 38505-1 | ISO and IEC | Principles for governing bodies on the use of data, applying the ISO/IEC 38500 governance model to data. Edition 2, published August 20, 2026, replaced the 2017 edition. |
| NIST CSF 2.0 | NIST | A cybersecurity framework, not a data governance one. ID.AM-07 asks for data inventories; ID.AM-08 asks that systems, software, services and data be managed throughout their life cycles. |
Pick one as the reference vocabulary, so teams argue about decisions, not terms.
The order of the work
- Pick the first use case. Choose a business problem that bad data is causing now: a regulatory report, a revenue figure that differs by system, an AI project waiting on clean data.
- Identify the critical data elements. List the few data elements that use case depends on. Microsoft Purview describes a critical data element as a logical grouping, for example one "Customer ID" that maps "CustID" in one table and "CID" in another.
- Name owners and stewards. An owner for each domain who decides, and a steward who does the daily work of definitions and quality issues. Write the decision rights down.
- Agree definitions. Build the business glossary for those elements only. Record who approved each definition and when.
- Set quality rules and measure them. Completeness, validity and consistency checks on each critical element, with a threshold the owner accepts. Publish the scores.
- Catalog and trace lineage. Register the source systems, the data sets and where each critical element flows.
- Set up the escalation path. A forum that meets on a schedule and settles disputes the stewards cannot.
- Repeat for the next use case. Add domains one at a time. Reuse the glossary and the rules.
Where to allow extra time
- Ownership. Agree the hours each owner will spend deciding, not only the title.
- Definitions. Two departments can each be right about a term that means different things in their systems. Settling it can change a reported figure, and someone has to accept that change.
- The catalog. Plan for descriptions, glossary links and their upkeep as ongoing work after go-live, not a one-time load.
- Fixing the data. Quality rules find problems. Agree in advance how the source system's team will fit corrections into its own backlog.
The detail plans leave out
- Data in SaaS applications. Customer, HR and finance data held by SaaS vendors needs an owner too, and access through the vendor's interfaces.
- Unstructured data. Include file shares, SharePoint and email in scope, not only databases. Microsoft Copilot surfaces organizational data to which a user has at least view permission, so the permissions on that content become a governance question.
- Access requests. Connect the catalog to an access request and approval path, so people can get the data they find.
- Change control. A new field, a renamed column or a new source can break a definition or a quality rule. Data changes need the same review as application changes.
- An exit for each data set. Governance covers the end of a data set's life as well as its start. Link each catalog entry to the retention policy.
The regulatory and cost angle
Several rules assume the inventory and ownership that governance produces.
- GDPR Article 30. Controllers must keep a record of processing activities: purposes, categories of data subjects and personal data, recipients, transfers to third countries, planned erasure time limits where possible, and a description of security measures. Organizations with fewer than 250 employees are exempt unless the processing is risky, not occasional, or involves special category or criminal data.
- California. Regulations from the California Privacy Protection Agency took effect January 1, 2026. Risk assessment compliance began that day, with attestations and summaries due to the agency by April 1, 2028. Cybersecurity audit certifications are due April 1, 2028 for businesses with revenue over $100 million, April 1, 2029 for $50 million to $100 million, and April 1, 2030 below $50 million.
- Banking. The Basel Committee's Principles for effective risk data aggregation and risk reporting, known as BCBS 239, were issued January 9, 2013, with global systemically important banks as their main target.
Beyond the tool, budget for the time of owners and stewards, the cleanup the quality rules uncover, and the change to reports when definitions are settled.
Skills and operations
A small central team runs the standards, the catalog and the reporting. Stewards sit in the business and need time in their objectives, not only a title. Data engineers implement the quality rules and lineage.
Catalog platforms now ship governance features. Microsoft Purview Unified Catalog organizes data into governance domains with data products, critical data elements, quality rules and health scores. Databricks Unity Catalog provides access control, lineage, audit logs and discovery for data and AI assets in Databricks, and has an open-source implementation.
Track the stewards' open issues, the share of critical elements with an approved definition, and quality scores against threshold. If the backlog only grows, revisit the scope or the time committed.
Settle these first
- Which business problem will the first phase fix, and by when?
- Who is the executive sponsor, and is it someone outside IT?
- Central, federated or domain-led: which model, and who has which decision rights?
- Which data elements are critical for the first use case?
- Which framework is the reference vocabulary?
- How much of each owner's and steward's time is committed, in writing?
Questions
- What is the difference between data governance and data management?
- Governance decides who owns data, what it means, what quality it needs and who may use it. Management is the technical work that carries those decisions out.
- Should data governance report to the CIO?
- It can, but the decisions belong to the business. IT runs the tools and controls. Business owners decide definitions and quality thresholds for their data.
- Do we need a data catalog before we start?
- No. Start with owners, definitions and quality rules for one use case. A catalog helps once there are decisions to record and data people need to find.
- What are critical data elements?
- The data elements a business process or report depends on, such as customer ID or revenue. Governing these first keeps the scope small enough to finish.
- Does NIST CSF 2.0 require a data inventory?
- CSF 2.0 can be adopted voluntarily or through government policies and mandates. It lists outcomes and does not prescribe how to achieve them. Its subcategory ID.AM-07 calls for inventories of data and metadata for designated data types, and ID.AM-08 calls for systems, software, services and data to be managed throughout their life cycles.