What is RACI for data governance roles?
Permalink to “What is RACI for data governance roles?”Data governance roles are going through a second reckoning. The first came from cloud and regulation. This one is coming from AI agents that now read and act on the same data these roles exist to protect. The core structure has not changed much: someone still owns the data, secures it, and uses it, but who does what is shifting fast, and two new specialist roles are joining the table.
Clear governance roles decide who sets the rules in the first place, which is why ownership cannot stay ambiguous. RACI matrices that map who is Responsible, Accountable, Consulted, and Informed help in removing that ambiguity.
Activity | CDAIO | Data owner | Data steward | Data product manager | Data custodian | Data user |
|---|---|---|---|---|---|---|
Setting governance policy and standards | A | R | C | C | I | I |
Classifying and certifying data domains | I | A | R | C | C | I |
Approving human and AI agent access | I | A | C | I | R | I |
Packaging data as a governed product | I | C | C | R/A | C | I |
Auditing and regulatory reporting | A | R | C | I | C | I |
Tip: Review your RACI matrix every 90 days and whenever major org or regulatory shifts occur.
Quick data governance role snapshots (what they actually do)
Permalink to “Quick data governance role snapshots (what they actually do)”- Chief data and AI officer (Accountable for the program): Sets the governance charter, owns the budget, and reports outcomes to the board.
- Data owner (Accountable for a domain): The senior business stakeholder for a specific area, such as customer or finance data. Approves classification, access policy, and resolves disputes within their domain.
- Data steward (Responsible for domain excellence): Bridges business and IT by defining metrics, enforcing quality rules, and reviewing the context an AI agent has drafted before it goes live.
- Data product manager (Responsible for the product): Packages governed, certified data as a supported asset with an owner, an SLA, and a roadmap, rather than a one-off export.
- Data custodian (Responsible for technical implementation): Runs encryption, tiered storage, backups, and access controls from a single console.
- Data user (Informed and empowered): Any employee or AI agent that turns approved data into value, and is expected to follow governance guardrails built into daily workflows.
Two specialized roles that are emerging as AI agents become mainstream:
- Context steward (Responsible for AI-context certification): Reviewing and certifying the context AI agents draft, rather than managing and protecting data alone
- AI governance lead (Consulted on regulatory compliance): Tracking regulation like the EU AI Act and translating it into governance controls
What skills do data governance responsibilities require?
Permalink to “What skills do data governance responsibilities require?”Right skill combinations determine whether governance enables or obstructs data-driven decisions. The skill mix is shifting too: reviewing AI-drafted work and interpreting regulation are now must-haves in roles where they used to be nice-to-haves.
Role | Must-have skills | Nice-to-have skills |
|---|---|---|
Chief data and AI officer | Storytelling, portfolio budgeting, change leadership | AI investment strategy, risk management |
Data owner | Domain expertise, policy sign-off, escalation ownership | Data literacy, budget authority |
Data steward | Domain expertise, glossary writing, stakeholder facilitation | Lightweight scripting (SQL, Python) |
Data product manager | Product road-mapping, SLA drafting, customer research | Agile coaching, pricing models |
Data custodian | Cloud security, IAM policies, backup and disaster recovery | Infrastructure as code, ABAC or RBAC tooling |
Data user | Analytical thinking, BI tools, basic SQL | Prompt engineering for GenAI |
Context steward | Reviewing AI-drafted context, semantic modeling, quality judgment | Familiarity with agent behavior, MCP or API concepts |
AI governance lead | Regulatory interpretation, risk classification, incident triage | Legal background, model documentation |
Build your data governance team the right way
Get 90-Day DG RoadmapHow many data governance roles do you need?
Permalink to “How many data governance roles do you need?”Right-sizing prevents both bottlenecks and governance theater, where roles exist without authority. Company size still drives most of the staffing decisions, but two new specialist roles only tend to appear once the AI footprint grows past a pilot.
Company size | Typical staffing pattern |
|---|---|
Fewer than 200 employees | 1 CDAIO (fractional), a few data owners as part-time "hats," shared stewards and custodians |
200 to 2,000 employees | Dedicated CDAIO, 3 to 5 data owners, 3 to 5 stewards, shared custodians, first data product manager hire |
More than 2,000 employees or regulated | Full team per domain: CDAIO plus council, data owners per domain, 5 to 10 stewards, 3 to 4 custodians, 1 data product manager per domain, plus a context steward and an AI governance lead |
Tip: Formalize roles only when workload or risk justifies it. Avoid premature hiring.
What tools do data governance roles need?
Permalink to “What tools do data governance roles need?”Without strong data governance, companies risk inaccurate insights, siloed systems, and privacy issues that can derail transformation efforts.
Tooling is a big part of why: 49% of organizations say inadequate tooling for automating data quality is the single biggest thing keeping their data from being trustworthy, according to Precisely’s 2025 planning insights research.
Traditional tools make this worse by spreading cataloging, quality, policy, and collaboration across disconnected platforms, so each role above ends up working from a different picture of the same data.
The fix the market has converged on is a single context layer: the infrastructure that stores, governs, and serves an organization’s business meaning to every person and every AI agent that needs it, rather than a data catalog or governance platform bolted onto a data warehouse.
- Chief data and AI officer: One dashboard for cross-domain risk and ROI reporting, with live compliance status instead of a quarterly slide deck.
- Data owner: A policy set once at the domain level that enforces automatically across every connected source.
- Data steward: Review and approval, not manual drafting. Atlan’s Context Agents draft descriptions, lineage, and quality context, so the steward certifies the work instead of writing it from scratch.
- Data product manager: A certified product catalog where governed datasets carry an SLA that other teams and agents can discover and reuse.
- Data custodian: One console for encryption, access, and audit logs across the existing stack, instead of five separate admin panels. Atlan’s Data Quality Studio runs checks directly inside Snowflake, Databricks, or BigQuery, so broken data gets flagged before an agent acts on it.
- Data user: A search interface that both people and AI agents can query through Atlan’s MCP server, under the same access rules either way.
Real stories from real customers: Governance at scale
Permalink to “Real stories from real customers: Governance at scale”"AI initiatives require more context than ever. Atlan's metadata lakehouse is configurable, intuitive, and able to scale to hundreds of millions of assets. As we're doing this, we're making life easier for data scientists and speeding up innovation."
— Andrew Reiskind, Chief Data Officer, Mastercard
Set up your data governance team with expert help
Book a Personalized Demo"Context is the differentiator. Atlan gave our teams the shared vocabulary and lineage to move from reactive data management to proactive AI enablement."
— Kiran Panja, Managing Director, Cloud and Data Engineering, CME Group
What are the biggest challenges with data governance roles, and how do you overcome them?
Permalink to “What are the biggest challenges with data governance roles, and how do you overcome them?”Most governance programs fail on role design rather than on tooling. Here are some of the biggest challenges, and fixes to overcome them.
Pitfall | Warning Signs | Impact | Solution |
|---|---|---|---|
Over-centralization | Request resolution >2 weeks, shadow IT solutions, "bureaucracy" complaints | User frustration, governance avoidance, policy violations | Federated model: 80% decisions at steward level, clear escalation paths |
Authority without accountability | Roles exist on charts but lack budget/decision rights | Governance theater, ignored policies, false security | Document RACI matrix, provide budget authority, tie to performance reviews |
Tool sprawl | Multiple data inventories, user confusion, no unified visibility | Information silos, duplicated effort, inconsistent enforcement | Unified platform or strong API integration between existing tools |
Governance theater | Meetings without decisions, acknowledged but unfollowed policies | Audit risk, cultural cynicism, wasted resources | Link performance to business outcomes, executive enforcement of decisions |
Role boundary confusion | Issues falling between roles, finger-pointing during incidents | Accountability gaps, duplicated work, slow resolution | Detailed RACI scenarios, regular cross-role collaboration, shared dashboards |
Stewards who expect to be automated away | AI skepticism and the fear of obsolescence in the face of full automation | Resistance to technology, slow adoption | Frame agents as removing the janitorial half of the job. The judgement and strategy stays with the stewards. |
Ungoverned agent identities | Agents are provisioned by engineering, inherit a service account, and never appear in the governance inventory | Unable to name the human accountable for an agent | Register agents as first-class consumers with an owner, an entitlement set, an expiry, and an audit trail |
Tip: Run a semi‑annual “tool audit” to retire redundant platforms before they create confusion.
How does AI transform data governance roles and responsibilities?
Permalink to “How does AI transform data governance roles and responsibilities?”The old model assumed a human steward doing a compliance job. The current one assumes an AI agent is about to take an action, and governance roles exist to tell it whether it’s allowed to do so. That’s a shift from human in the loop to human on the loop: agents do the volume work, stewards approve.
Governing the AI is not really about governing the AI. To govern the AI you need to govern the context it reads. This is reflected in the following trends shaping data governance roles:
- A new context steward role is emerging. Gartner describes stewards progressing along a maturity curve (AI-skeptic, AI-assisted, AI-augmented, AI-native), with the context steward focused on how data is understood and used, not just how it’s managed and protected.
- Agents draft, stewards approve. Context Agents author the descriptions, glossary terms, and metadata that used to take stewards months to write by hand; the steward’s job shifts to reviewing and certifying that work, not producing it from scratch.
- Custodians monitor context, not just infrastructure. Lineage and quality signals stream into the tools teams already use, so a custodian sees drift the moment it happens instead of during a quarterly audit.
- Data Product Managers package certified context as reusable assets. Governed, certified data products become the node other teams and agents build on, rather than one-off deliverables.
- A regulation-watching role becomes necessary. High-risk AI systems now carry specific data governance obligations under the EU AI Act, including data quality, traceability, and protections that apply even at inference time. Naming an AI governance lead, sometimes called an AI officer, is not strictly mandated for every company, but it is strongly recommended for any organization deploying high-risk AI.
Ready to define your data governance team?
Permalink to “Ready to define your data governance team?”The core structure of data governance has not changed. Someone still owns the data, someone secures it, someone defines it, and someone uses it. What changed is that the fourth category now includes software that acts without asking. That reframes role definitions.
Instead of retrofitting roles under pressure, make sure you name the accountable human for every agent in production, move stewards from authoring to certifying, and express policy where agents can read it.
It’s also important to reiterate that while each team member has a distinct role, these data governance roles depend on each other. Everyone must collaborate effectively to help their organization achieve its business and data goals.
Set up your data governance team with expert help
Book a Personalized DemoFAQs about data governance roles and responsibilities
Permalink to “FAQs about data governance roles and responsibilities”1. What is a data governance role?
Permalink to “1. What is a data governance role?”A data governance role refers to a specific position within an organization responsible for managing, protecting, and ensuring the quality of data. Key roles include data admin, data steward, data custodian, and data user, each with distinct responsibilities that contribute to effective data governance.
2. Who should be in charge of data governance?
Permalink to “2. Who should be in charge of data governance?”Data governance should ideally be overseen by a dedicated data governance team, which may include a Chief Data Officer (CDO) or a data governance committee. This team is responsible for establishing policies, standards, and practices that ensure effective data management across the organization.
3. Can one person handle multiple data governance roles?
Permalink to “3. Can one person handle multiple data governance roles?”Yes, especially in smaller organizations. Senior data professionals often combine Administrator and Custodian duties, while business analysts serve as both Stewards and Users. Key is documenting which responsibility applies to each decision.
4. How do governance roles differ from data engineering roles?
Permalink to “4. How do governance roles differ from data engineering roles?”Governance focuses on policy, quality, compliance while engineering builds technical infrastructure. Governance provides “what and why” through business context, engineering delivers “how” through technical implementation.
5. What is the difference between a data owner and a data steward?
Permalink to “5. What is the difference between a data owner and a data steward?”The owner is accountable and the steward is responsible. An owner is usually a business leader who signs off on how a domain’s data may be used and answers for the consequences, while a steward is the practitioner who writes definitions, sets quality thresholds, certifies assets, and resolves day-to-day conflicts. In smaller organizations one person often holds both, which is workable as long as the documentation states which authority is being exercised in a given decision.
6. What’s the biggest governance implementation mistake?
Permalink to “6. What’s the biggest governance implementation mistake?”Creating accountability without authority. Roles need decision power, budget allocation, executive backing to drive change. Governance theater—roles existing only for audits, increases risk through false confidence.
7. How often should data governance role definitions be updated?
Permalink to “7. How often should data governance role definitions be updated?”Annually or during major changes: new regulations, technology adoption, organizational restructuring. Programs deploying agentic AI generally move to a semi-annual or quarterly cadence, because the set of things an agent can reach changes faster than an annual review cycle can track.
8. How does AI change data stewardship?
Permalink to “8. How does AI change data stewardship?”The steward’s output shifts from production to certification. Agents can now draft descriptions, glossary terms, metrics, and quality rules at a volume no human team could match, so the scarce resource becomes judgement rather than typing. The role also expands to cover a new consumer: stewards increasingly certify context that AI agents read at inference, which means their approval directly determines whether an autonomous system acts on a correct definition or a plausible-sounding wrong one.
Sources
Permalink to “Sources”Precisely | Data Integrity | 2025 Planning Insights: Data Quality Remains the Top Data Integrity Challenge https://www.precisely.com/data-integrity/2025-planning-insights-data-quality-remains-the-top-data-integrity-challenges/
Precisely | Data Integrity | 2025 Planning Insights: Data Governance Adoption Has Risen Dramatically https://www.precisely.com/blog/data-integrity/2025-planning-insights-data-governance-adoption-has-risen-dramatically
