To scale data governance across business domains, keep a small central team that owns the few rules every domain must share, and give everything else to named owners and stewards inside each domain. Roll out domains in small waves, starting where business value is highest, and let automation carry shared standards such as classification and access policy into every new domain. Then measure each domain separately, so you see which one is drifting before it slows the whole program.
Why does data governance stop scaling after the first domain?
In the first domain, a central team writes the glossary, tags sensitive columns and approves access by hand, and keeps up. By the third or fourth domain, two problems appear.
The first is a queue. Every definition, classification and access decision still routes through the same few people, so requests wait and teams work around the process. The second is drift. Domains tired of waiting write their own rules, and soon “customer” has three definitions that no cross-domain report can reconcile.
Both come from the operating model, and more central headcount only delays them. Gartner predicts that 80% of data and analytics governance initiatives will fail by 2027 because they lack a real or manufactured crisis to drive them. In September 2026 it added that 60% of organizations that ignore data governance culture challenges will fail to govern AI successfully by 2027.
Governance scales when the people closest to the data own it, for an outcome they care about. For domain definitions and boundary types, see data domain data governance.
Which operating model scales data governance across business domains?
In the Enterprise Data Strategy Board’s 2025 survey of Fortune 1000 data leaders, 36% ran centralized governance, 36% federated, and 29% a hybrid.
| Model | Who decides | Where it breaks | When it fits |
|---|---|---|---|
| Centralized | One central team writes and enforces every rule | Becomes a queue as request volume grows | One or two domains, early programs |
| Federated | A central body sets shared rules; domains decide and execute locally | Slides into decentralized if shared rules are not enforced automatically | Three or more domains with their own data teams |
| Decentralized | Each domain sets and enforces its own rules | Definitions diverge and cross-domain reporting fails | Independent units that rarely share data |
To scale across domains, most programs move to federated. Zhamak Dehghani, who defined data mesh, calls the version that holds at scale federated computational governance: domain data product owners and platform owners share decisions, and domains keep local decision power while following a set of global rules. “Computational” means the platform executes the global rules, so no domain has to remember them.
Hybrid works as a transition if the plan says which decisions move to domains, and when. The data governance models page compares the three in more depth.
The AI Context Stack
A brief on the layers an agent needs from enterprise data, from metadata and lineage to definitions and access rules.
Get the AI Context StackWhat should the central team decide, and what should each domain decide?
One test settles most federated decisions: if a decision affects more than one domain, the center or the council owns it. If it affects one domain, that domain owns it.
| Decision | Central team or council | Each domain |
|---|---|---|
| Sensitivity classes and PII rules | Defines the classes and what each requires | Applies them to its own assets |
| Access policy | Sets the rule for who may see restricted data | Approves requests for its own data within that rule |
| Business glossary | Owns terms that cross domains, such as “customer” | Owns terms used only inside the domain |
| Data model | Sets naming and identifier standards for shared keys | Designs its own models and tables |
| Data quality | Sets how quality is expressed and reported | Sets thresholds for its own data products |
In her description of a federated mesh, each domain decides its data model and how to measure its quality, while the federation sets a common way to identify the same listener across domains and a common standard for modeling quality and service-level objectives. Shared identifiers and a shared quality vocabulary are what make data usable across domains.
Publish this table before the second domain goes live.
Who does what in each business domain?
McKinsey describes a typical governance structure with three components: a central data management office, governance roles organized by domain, and a data council that links data strategy to corporate strategy, approves funding and resolves issues.
Inside each domain, an executive owner answers for the domain; in McKinsey’s example, the CFO and CMO each owned several. A data owner signs off on classification and access and settles disputes. Stewards write definitions, set quality thresholds and certify assets, and data engineers build pipelines and apply technical controls. The central office is a small team that writes shared standards, runs the platform and trains new domains. The council, made of domain owners plus the central team, meets to settle cross-domain conflicts, approve policy changes and review each domain’s progress.
Name a person for every role before a domain goes live, and keep stewards inside the business, because the people who use the data know first when a definition is wrong. Sizing guidance is on the data governance roles page.
How do you roll governance out from one domain to many?
Each wave should be faster to onboard than the one before it.
- Rank domains by value and readiness, starting where governed data changes a business decision and an owner is ready. McKinsey advises starting with two to three domains, each fully functional within several months; in its example, leaders worked in sprints and checked in with the CEO every few weeks.
- Treat the first wave as the template. Record its roles, glossary format, classification steps, access workflow and quality checks as an onboarding kit.
- Onboard later waves with the kit. The central team coaches and the domain does the work; if a domain still needs the center to do its work, fix the kit.
- When a second domain needs a term the first one defined, move it to the enterprise business glossary and give it to the council.
- After every wave, automate the step that slowed it most, usually access approvals or classification.
Start a new wave only when every domain in the previous wave runs without daily central help and its owner reports its own metrics. Starting earlier rebuilds the queue.
Context Gap Calculator
Estimate how much of your data estate is missing the context an AI agent needs to answer correctly.
Calculate Your Context GapHow does automation keep shared standards consistent across domains?
Federated governance holds only if global rules apply themselves. Dehghani’s goal is the automated execution of decisions by the platform, because a rule each team must remember gets applied differently in each domain.
Four kinds of automation carry most of the load:
- Classification that travels with the data. Tag a column as PII once, and lineage carries the tag into every downstream table, dashboard and data product.
- Policy as code. Access rules are written once, versioned centrally and evaluated at query time. The domain owner approves who gets in, and the rule decides what they see.
- Data contracts. A contract is a machine-readable agreement between a producer and its consumers. The Bitol project’s Open Data Contract Standard defines 15 sections for a YAML contract, including schema, data quality, roles and service-level agreement. With one format everywhere, consumers can trust another domain’s product without a meeting.
- Audit logs as a by-product. Each automated decision is logged, so “who saw restricted data?” becomes a query.
Automation moves the central team from reviewing requests to maintaining rules and templates, so a small team can support many domains. See automated data governance for more.
How do you measure whether data governance is scaling across domains?
Score every domain on the same metrics; one enterprise number hides the laggard.
| What to measure | Example metric | What it tells you |
|---|---|---|
| Coverage | Share of critical assets with a named owner and steward | Whether accountability exists |
| Consistency | Cross-domain terms with more than one active definition | Whether drift has started |
| Speed | Median time from access request to decision | Whether governance has become a queue |
| Onboarding | Time to a new domain’s first governed data product | Whether onboarding gets faster each wave |
| Enforcement | Share of policy checks run automatically | Whether shared rules apply themselves |
| Adoption | Active users per domain who search, certify or define assets | Whether people use governed data |
Report each domain’s trend against its go-live baseline at every data council meeting. Rising access time means a domain needs more stewards or automation. A rising count of conflicting definitions means terms that should be shared are still owned locally.
In a March 2026 Gartner survey of 223 D&A leaders, cultural resistance outweighed funding constraints as the main reason governance initiatives fail, 60% to 40%. A domain with full coverage and low adoption is governed on paper only.
How does Atlan help scale data governance across business domains?
In Atlan, domains, data products and data contracts are first-class objects in the catalog. Each domain controls its own products, people and policies, and the central team keeps shared standards across all of them.
Autodesk shows how that plays out. Its Analytics Data Platform team adopted a data mesh so any team could publish its own data, and through Atlan 60 domain teams now have full visibility into the consumption of their data products. Mark Kidwell, Chief Data Architect, Data Platforms and Services, said: “We don’t scale this by scaling up the core team. We had to enable other teams to do all these things.”
Each publishing team named a product owner and data engineers, and the platform team supplied the tooling, standards and enablement to one team after another. Kidwell counts 45 use cases through the self-service initiative since 2021. Atlan integrates with Snowflake out of the box for Autodesk, and Kidwell describes its role this way: “Atlan is the layer that brings a lot of the metadata that publishers provide to the consumers, and it’s where consumers can discover and use the data they need.”
Atlan in Action: Live Context Layer Demos
Watch Atlan's context layer in a live demo session.
Watch the Live DemoFAQs about scaling data governance across business domains
What is the difference between domain governance and federated governance?
Domain governance assigns data to business domains and gives each domain an owner. Federated governance is the operating model that lets many domains work together: a central body sets shared rules for classification, access and cross-domain terms, and each domain applies them locally. Domains without federation tend to drift apart as each writes its own rules.
Who resolves definition conflicts between business domains?
The data council, which brings domain owners and the central team together. When two domains define the same term differently, the council picks one enterprise definition, assigns an owner, and moves the term to the shared glossary. Terms used inside a single domain stay with that domain’s stewards.
Do you need a data mesh to scale governance across domains?
No. Data mesh is one architecture that uses federated governance, and the same model works on a central warehouse or lakehouse. What matters is a clear split of decisions between the center and the domains, named owners and stewards in each domain, and automation that applies shared rules without manual review.
How does scaling governance across domains affect AI agents?
An agent answering a question that spans two domains needs one definition of each shared term and the same access rules in both. If finance and sales define revenue differently, the agent returns a confident wrong answer. A shared glossary, consistent classification and policy enforced at query time keep agents accurate across domains.
Sources
- Gartner Predicts 80% of D&A Governance Initiatives Will Fail by 2027, Gartner
- Gartner Predicts 60% of Organizations That Ignore Data Governance Culture Challenges Will Fail to Govern AI Successfully by 2027, Gartner
- 2025 State of Enterprise Data Governance, Enterprise Data Strategy Board
- Data Mesh Principles and Logical Architecture, martinfowler.com
- Designing Data Governance That Delivers Value, McKinsey
- Open Data Contract Standard, Bitol
