--- title: "How to Assign Owners and Stewards for AI-Ready Data, Per Use Case" url: "https://atlan.com/know/ai-agent/ai-ready-data/ai-ready-data-owners-and-stewards/" description: "How to assign owners and stewards for AI-ready data per use case: AI drafts context, stewards check it, owners certify it. Seven steps, tables, and pitfalls." author: "Emily Winks" author_role: "Data Governance Expert" published: "2026-11-06" updated: "2026-11-06" --- ## How do you assign owners and stewards for AI-ready data? Assign them per AI use case, not once per domain. Name the agent and its risk tier, map the assets it reads, and give each domain an accountable owner, a backup, and a steward. AI drafts the missing descriptions, the steward checks them, and the owner certifies the assets and records that named use in an approved-uses field before the agent goes live. Then route changes through approval workflows, lock down bulk and agent writes, and recertify when sources or eval scores change. The 7 steps: - **Name the use case** and its risk tier - **Map the assets** and domains the agent reads - **Assign roles:** owner, backup, and steward per domain - **Draft context** with AI, then have the steward check it - **Certify** fitness for the named use - **Route changes** through approval workflows - **Deliver and recertify** on set triggers ## Key takeaways - Assign owners per AI use case: the owner certifies that the context an agent receives fits that one job. - Stewards review every AI draft before an owner signs off. The agent's harness reads only what the owner approved. - An Atlan certificate is a single status, so list each approved use in a custom metadata field. - Approval workflows cover UI edits; metadata policies apply on every write path, so set both. --- Assign owners and stewards per AI use case, on top of domain ownership: the owner certifies that the meaning, lineage, and access rules an agent receives fit that one job. In Atlan's context layer, **Context Agents** draft missing descriptions, stewards check them, and the owner certifies the assets and records the approved use, which becomes the release gate once retrieval checks it. As [Gartner's guidance on AI-ready data](https://www.gartner.com/en/articles/ai-ready-data) puts it, "There is no way to make data AI-ready in general or in advance. The readiness of data for AI depends on how the data will be used." ### Score Your Governance Readiness First Answer questions across five dimensions, from people and ownership to policy and outcomes. It returns a score, a maturity band, and a sequenced 90-day plan. [Read the skill](/skills/governance-readiness-score.md). *Paste into a new chat* ``` Use the skill at https://atlan.com/skills/governance-readiness-score.md to score our data and AI governance readiness, including ownership and stewardship, before we assign owners for an AI use case. Ask me for whatever it needs. ``` *Run once in a terminal* ``` curl -fsSL --create-dirs \ -o ~/.agents/skills/governance-readiness-score/SKILL.md \ https://atlan.com/skills/governance-readiness-score.md ``` *For an agent* ``` curl -fsSL https://atlan.com/skills/governance-readiness-score.md ``` Human Agent ClaudeChatGPTGeminiCursorOther Copy --- A BI analyst used to carry this context: which revenue table excluded refunds, and who to call when a number looked off. An agent needs it written down beside the data, with a named person behind it. **Effort:** heaviest for the first use case, lighter after. **Difficulty:** intermediate. | Step | Who does it | |---|---| | 1. Name the use case and risk tier | Use-case owner, sponsor | | 2. Map assets and domains | Steward | | 3. Assign owner, backup, steward | Domain lead | | 4. Draft context | Context Agents, then steward | | 5. Certify for the use | Domain owners; use-case owner for the package | | 6. Route changes, lock writes | Governance admin | | 7. Deliver and recertify | Owners, agent harness team | Draft, check, decide: the owner certifies for one named use, the agent reads certified context, and recertify triggers loop back. --- ## Why do AI use cases need their own owners and stewards? AI use cases need their own sign-off because an agent reads data without the analyst who used to explain it. Domain ownership gives a table an owner and escalation path, and [data contracts for AI](https://atlan.com/know/ai-agent/data-for-ai/data-contracts-for-ai/) pin down schema and freshness. What is usually missing is a fitness decision for one consumer (this table, with this definition, suits this decision) stored where an agent reads it. According to [Gartner (2026)](https://www.gartner.com/en/newsroom/press-releases/2026-05-11-gartner-says-lack-of-semantics-causes-inaccurate-artificial-intelligence-agents-and-wasted-spending), agents without context (the relationships and rules in an organization's data) cannot operate accurately, and D&A leaders should "establish a context layer as a core component of their D&A infrastructure." Somebody has to decide which context is correct, so the [properties that make data AI-ready](https://atlan.com/know/ai-agent/data-for-ai/what-makes-data-ai-ready/) include an owner; without one, [AI projects stall on data rather than models](https://atlan.com/know/ai-readiness/why-do-ai-projects-stall-on-data-rather-than-models/). For high-risk AI systems, [Article 10 of the EU AI Act (European Union, 2024)](https://artificialintelligenceact.eu/article/10/) holds training, validation, and testing data to governance "appropriate for the intended purpose of the high-risk AI system" (see the [EU AI Act summary](https://atlan.com/know/ai-readiness/the-eu-ai-act-summary/)). Its test, readiness judged against a stated purpose, fits run-time context too. A certificate with no named use tells people which table is canonical; the owner's named use is what says it fits an agent's job. That is the gap between [AI-ready and analytics-ready data](https://atlan.com/know/ai-agent/ai-ready-data/ai-ready-data-vs-analytics-ready-data/): the analyst carried the context, and now the owner records it. --- ## What do you need before assigning owners and stewards? Five conditions decide whether assignments hold. According to [Gartner (2025)](https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk), 63% of organizations either do not have or are unsure if they have the right data management practices for AI. | Prerequisite | How to check | |---|---| | Named AI use case | The agent's decision fits in one sentence | | Executive accountability for AI risk | A named executive owns the tier policy | | Domain list | Every in-scope asset maps to a domain | | Owners, certificates, approved-uses field | All three show on assets the agent reads | | Steward time | Hours named in the sprint plan | The [NIST AI RMF Core (NIST, 2023)](https://airc.nist.gov/airmf-resources/airmf/5-sec-core/) puts AI risk decisions on executive leadership (GOVERN 2.3) without prescribing titles, so treat roles as decision rights. Settle [who owns AI governance](https://atlan.com/know/ai-agent/who-owns-ai-governance-caio-vs-cdo-vs-cto/) first; an [AI readiness assessment](https://atlan.com/know/ai-agent/data-for-ai/ai-readiness-assessment/) and the [AI readiness guide](https://atlan.com/know/ai-readiness/ai-ready-data/) find other gaps. --- ## Step 1: Name the AI use case and its risk tier Every later step is judged against the use-case record written here. 1. Write down the agent and the decision it informs: `churn-forecast-agent` informs finance's retention budget. 2. Name the use-case owner, usually the AI product owner, and assign a risk tier. 3. Record what "good enough" means. According to [McKinsey (2026)](https://www.mckinsey.com/capabilities/mckinsey-technology/our-insights/ai-data-readiness-the-key-to-scaling-impact), "The objective is not to achieve perfect data before starting, but to define what "good enough" means for each use case based on business needs and the risk profile of the data and related business processes." | Risk tier | Example use | Who signs off | Recertify when | |---|---|---|---| | High (regulated, customer-facing) | Credit or claims agent | Domain and use-case owners; executive signs tier; compliance and privacy review | Schema or definition change; fixed date | | Medium (internal decisions) | Forecasting agent | Domain and use-case owners | Schema or metric change; quarterly | | Low (exploratory Q&A) | Internal [talk-to-data](https://atlan.com/know/ai-agent/ai-ready-data/ai-ready-data-for-talk-to-data/) pilot | Domain owner, lighter evidence | Eval scores below baseline | Map them to your [AI risk management](https://atlan.com/know/ai-readiness/ai-risk-management/) categories. Starting with one use case, as in a [context layer pilot](https://atlan.com/know/ai-agent/context-layer/how-to-start-a-context-layer-pilot-or-proof-of-value/), keeps the first round finishable. --- ## Step 2: Map the assets and domains the use case touches This step bounds the use case to a list of assets, each tagged to a domain. 1. List the tables, metrics, glossary terms, and documents the agent will read. 2. Trace each one upstream with [data lineage for AI](https://atlan.com/know/ai-agent/data-for-ai/data-lineage-for-ai/): to the system of record for a high-tier use, at least one hop for a low-tier one. 3. Tag every asset to a domain and note the access rules that travel with it. Example: `churn-forecast-agent` reads `fct_revenue_daily` (finance domain, fed by billing) and `dim_customer` (customer domain, fed by the CRM). Two owners sign their assets; the use-case owner signs for the join. Keep ownership and access apart. Atlan's [access design guidance (Atlan docs)](https://docs.atlan.com/product/capabilities/governance/access-control/best-practices/design-access-and-personas) says "Domains express ownership and organization; personas express access." [Role-based access control in a context platform](https://atlan.com/know/ai-agent/context-layer/context-layer-role-based-access-control/) handles access; both belong in [governance rules readable by AI agents](https://atlan.com/know/data-governance/why-should-data-governance-rules-be-readable-by-ai-agents/). [Extending your data catalog for AI readiness](https://atlan.com/know/ai-agent/ai-ready-data/extend-data-catalog-for-ai-readiness/) adds this use-case layer. --- ## Step 3: Assign an accountable owner, a backup, and a steward per domain Each domain the use case touches gets an accountable owner, a backup, and a steward with booked hours, plus a forum with an executive tiebreak for cross-domain conflicts. Role definitions live in [data steward vs data owner](https://atlan.com/data-steward-vs-data-owner/). In Atlan, owners can be users or groups, per [Add owners (Atlan docs)](https://docs.atlan.com/product/capabilities/discovery/how-tos/add-owners); name one individual as accountable and use groups for routing. Pick the owner from the business. Oliwia Berdak, VP, Research Director at Forrester: "While technology teams often lead AI efforts, business stakeholders must define the problems that AI is solving." ([Forrester, 2025](https://www.forrester.com/blogs/strategic-ai-readiness-how-to-move-from-hype-to-scalable-impact/)) Who signs what: | Task | AI | Steward | Domain owner | Use-case owner | |---|---|---|---|---| | Write missing descriptions | Drafts | Reviews, edits, rejects | Spot-checks | Informed | | Confirm lineage and freshness | Not delegated | Checks | Accepts the check | Informed | | Define meaning and permitted use | Not delegated | Proposes | Decides | Consulted | | Certify assets for the use | Not delegated | Prepares evidence | Certifies | Informed | | Accept the assembled context for the agent | Not delegated | Prepares | Consulted | Accepts | An [AI center of excellence](https://atlan.com/know/ai-agent/ai-agent-governance/how-to-build-ai-center-of-excellence/) settles who defines "active customer." See [scaling governance across business domains](https://atlan.com/know/data-governance/how-to-scale-data-governance-across-business-domains/), [who should own the context layer](https://atlan.com/know/context-layer-ownership-data-vs-ai-teams/), and [data governance roles and responsibilities](https://atlan.com/data-governance-roles-and-responsibilities/). --- ## Step 4: Let AI draft the missing context, then have the steward check it The goal is for owners to approve text rather than write it. AI fills the empty descriptions, and a steward reviews each one. Atlan's two drafting routes behave differently: - [Context agents](https://atlan.com/know/context-agents/), run by an Admin or Governance Admin, enrich only assets missing the target attribute, and "Text a person wrote is never touched," per the [metadata enrichment FAQ (Atlan docs)](https://docs.atlan.com/product/capabilities/governance/context-agents-studio/faq/metadata-enrichment). - Single-asset Atlan AI suggestions go live when a user clicks Apply; bulk column runs fill only empty columns. Atlan AI changes are marked "Updated using Atlan AI" in the activity log, per [Use Atlan AI for documentation (Atlan docs)](https://docs.atlan.com/product/capabilities/atlan-ai/how-tos/use-atlan-ai-for-documentation). AI text can land on an already certified asset, so run the steward queue from those activity-log entries and keep bulk AI runs off certified in-scope assets (Step 6). Done means in-scope assets carry grain, keys, the agreed definition, and exclusions, and anything unreviewed stays at Draft. Start with [KPI definitions and a business glossary](https://atlan.com/know/ai-agent/how-to-manage-kpi-definitions-and-a-business-glossary-in-a-data-catalog/); see also [trust and control with context agents](https://atlan.com/know/fear-trust-control-context-agents/), the [AI data steward](https://atlan.com/ai-data-steward/), and [automated data stewardship](https://atlan.com/know/data-governance/automated-data-stewardship/). --- ## Step 5: Have the owner certify fitness for the named use case This step is the release gate: domain owners certify their assets for the named use, and the use-case owner accepts the package. It holds only if the harness checks it (Step 7). Atlan certificates are Verified, Draft, Deprecated, or none, with an optional message, per [Add certificates (Atlan docs)](https://docs.atlan.com/product/capabilities/discovery/how-tos/add-certificates). The per-use answer goes beside that single status: a custom metadata property of the Options type with **Allow multiple values** on, per [Manage custom metadata structures (Atlan docs)](https://docs.atlan.com/product/capabilities/governance/custom-metadata/how-tos/manage-custom-metadata-structures), listing every approved use: ```text Certificate: Verified Approved AI uses: churn-forecast-agent (tier: medium) Message: Owner: finance domain. Steward check 2026-11-04: lineage, freshness, descriptions complete. Recertify by 2027-02-01 or on any schema change to fct_revenue_daily. ``` A scheduled check reads the recertify date from the message. For a fuller record, pick the container that fits: | Where the decision lives | Use it when | |---|---| | Certificate, message, and approved-uses field | A few agents share the asset | | Context repository approved in **Context Engineering Studio** | You package context for one agent | | Data product | Many consumers read one domain's curated assets | The studio proposes assets for one use case with a rationale for each, a person approves, and it assembles the [context repository for AI agents](https://atlan.com/know/ai-agent/context-repository-for-ai-agents/), per [the Atlan docs](https://docs.atlan.com/product/capabilities/governance/context-engineering-studio/concepts/what-is-context-engineering-studio). Make the use-case owner that approver: it is the one sign-off covering the combination of assets, the [context engineering](https://atlan.com/know/what-is-context-engineering/) step no domain owner can judge alone. [Packaging data products in a catalog](https://atlan.com/know/ai-agent/data-for-ai/how-to-package-data-products-data-catalog/) covers the third option. Run the owner through the [AI-ready data checklist](https://atlan.com/know/ai-agent/data-for-ai/ai-ready-data-checklist/) first; a scoped, signed approval is what [trusted data for AI](https://atlan.com/know/ai-readiness/what-is-trusted-data-for-ai-and-how-do-you-get-it/) looks like in practice. --- ## Step 6: Route changes through approval workflows and close the other write paths Certified context should not change without the owner knowing, and the agent should not edit its own gate. Per [Automate data governance (Atlan docs)](https://docs.atlan.com/product/capabilities/governance/stewardship/how-tos/automate-data-governance), change management covers descriptions (by people or Atlan AI), certificates, owners, and custom metadata, but workflows trigger only for edits in the Atlan UI. Bulk updates through playbooks, Atlan AI, the API, SDK, and CLI, spreadsheet add-ins, and the Asset Import app skip them. Metadata policies apply on every path, since each change runs as the user who makes it. When asset owners approve, [Create governance workflows (Atlan docs)](https://docs.atlan.com/product/capabilities/governance/stewardship/how-tos/create-governance-workflows) requires at least one other approver. Close the gaps: 1. In [metadata policies (Atlan docs)](https://docs.atlan.com/product/capabilities/governance/access-control/metadata-policy), grant Update on in-scope assets only to their owners and stewards. Certification status and ownership sit under the Update permission. 2. Run the agent under an identity with Read but no Update on those assets; Atlan MCP respects Atlan permissions, so an identity without Update cannot change certification through it. 3. Audit the activity log for non-UI changes and send in-scope changes back through Step 5. Test it: change an in-scope certificate in the UI and confirm an approval request appears; repeat through a playbook, the SDK, and the agent's MCP identity, and confirm each is refused or logged. Keep approvals where owners work, as in [embedded data governance](https://atlan.com/know/data-governance/embedded-data-governance/), with [AI agent governance](https://atlan.com/know/ai-agent-governance/) on top (see [data governance vs AI governance](https://atlan.com/know/data-governance-vs-ai-governance/)). --- ## Step 7: Deliver certified context to the agent and recertify on triggers The owner's decision counts only once it reaches the agent. Atlan MCP is "a hosted server, available on every Atlan tenant with nothing to install," and "every action respects the permissions you've already set in Atlan," per the [Atlan MCP overview (Atlan docs)](https://docs.atlan.com/product/capabilities/atlan-ai/how-tos/atlan-mcp-overview). Per the [MCP tools reference (Atlan docs)](https://docs.atlan.com/product/capabilities/atlan-ai/references/mcp-tools), `search_assets` filters on certification status, `get_assets` returns owners and custom attributes, and write tools such as `update_assets` wait for approval of a preview. Atlan supplies the record; your harness enforces it and fails closed: ```text for each asset the agent plans to read: if certificate != Verified -> skip, log "not certified" if use_case not in approved_ai_uses -> skip, log "not approved for this use" otherwise read it, and keep the owner from get_assets if a required asset was skipped -> stop, and return the owner to ask ``` That filter lives in the [AI agent harness](https://atlan.com/know/how-to-build-ai-agent-harness/) and [Atlan MCP server](https://atlan.com/know/what-is-atlan-mcp/) setup; see [AI-ready data over MCP](https://atlan.com/know/ai-agent/ai-ready-data/ai-ready-data-mcp/) and [MCP and the semantic layer](https://atlan.com/know/ai-agent/semantic-layer/mcp-semantic-layer-for-ai-agents/). Recertify on an upstream schema or contract change (alert the steward), a metric definition change (domain owner), eval scores below the baseline set at certification (use-case owner; see [context evals vs data quality testing](https://atlan.com/know/ai-agent/ai-agent-evals/context-evals-vs-data-quality-testing/)), or the date in the message, but not routine loads. --- ## What goes wrong when you assign owners and stewards for AI? Six failures recur, each with a fix. | Mistake | Fix | |---|---| | Ownership by domain only | Name the use, then certify against it | | AI drafts treated as approved | Steward queue; bulk AI runs off certified assets | | One owner reviewing every agent | Tier by risk | | Certification with no end date | Recertify triggers and a scheduled date check | | Agent identity can write metadata | Read-only identity for the agent | | No cross-domain arbitration | Name the council and the tiebreak | Plan for reviewer fatigue: AI drafts at volume, so let the tier set review depth. Metric owners are hardest, because a disputed definition in the [semantic layer](https://atlan.com/know/semantic-layer/) breaks every agent that reads it, a common way [governance implementations fail](https://atlan.com/know/ai-agent/ai-agent-governance/why-data-governance-implementations-fail/). --- ## How does Atlan support owner and steward sign-off for AI? Atlan keeps owners, certificates, approvals, and agent delivery in one context layer, so the agent reads the decision a person made. In Atlan's [enterprise context layer](https://atlan.com/know/what-is-the-enterprise-context-layer/): - Context Agents draft only missing descriptions; Atlan AI changes are marked in the activity log. - Owners certify assets, and a custom metadata property lists the approved uses. - Workflows route UI changes; metadata policies govern every write path. - Context Engineering Studio builds a use case's context repository after a person approves it. - Atlan MCP filters by certification, returns owners and custom attributes, and respects Atlan permissions. A [data catalog and a context layer](https://atlan.com/know/data-catalog-vs-context-layer/) part ways here: the catalog records who owns a table, and the [context layer for enterprise AI](https://atlan.com/know/context-layer-enterprise-ai/) carries that owner's per-use decision to the agent. --- ## What should you do after owners and stewards are in place? After the first use case, track the share of in-scope assets Verified with the use listed, days from draft to certification, and open escalations. The next use case reuses named owners and existing evidence, then gets its own decision: the owner adds it to the approved-uses field only after checking the assets against it. [Building AI-ready data from day zero](https://atlan.com/know/ai-agent/ai-ready-data/build-ai-ready-data-from-day-zero/) and [SAP and ERP data for agents](https://atlan.com/know/ai-agent/ai-ready-data/ai-ready-sap-erp-data-for-agents/) apply the pattern to new and legacy estates; the [AI-ready data platforms](https://atlan.com/know/ai-agent/data-for-ai/best-ai-ready-data-platforms/) list compares the tooling. If an owner cannot name the agents their approval covers, start the next round there. --- ## FAQs about assigning owners and stewards for AI-ready data ### 1. What is the difference between a data owner and a data steward for AI? The owner decides what the data means and which uses are permitted, then certifies it for a named AI use. The steward checks lineage, freshness, and AI drafts, and prepares the evidence. ### 2. Who should be assigned as the data owner for an AI use case? The business lead accountable for the decision the data supports, in the domain that produces it, plus a named backup. IT runs the platform; the business holds the sign-off. ### 3. How do you certify data for AI use? Name the use case, have the steward check lineage and freshness for it, then have the owner set the asset to Verified and add the use to an approved-uses field. ### 4. Can AI replace data stewards? No. AI drafts missing descriptions; stewards check them and owners decide meaning and permitted use. Context agents in Atlan leave text a person wrote untouched. ### 5. Can one data asset be certified for one agent and not another? Yes. An Atlan certificate is a single status, so list each approved use in a multi-value custom metadata field, or approve a context repository for that one use. --- ## Sources 1. [Lack of AI-Ready Data Puts AI Projects at Risk, Gartner (26 February 2025)](https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk) 2. [AI-Ready Data Essentials to Capture AI Value, Gartner](https://www.gartner.com/en/articles/ai-ready-data) 3. [Gartner Says Lack of Semantics Causes Inaccurate AI Agents and Wasted Spending, Gartner (11 May 2026)](https://www.gartner.com/en/newsroom/press-releases/2026-05-11-gartner-says-lack-of-semantics-causes-inaccurate-artificial-intelligence-agents-and-wasted-spending) 4. [AI data readiness: The key to scaling impact, McKinsey (23 June 2026)](https://www.mckinsey.com/capabilities/mckinsey-technology/our-insights/ai-data-readiness-the-key-to-scaling-impact) 5. [Article 10: Data and Data Governance, EU Artificial Intelligence Act (2024)](https://artificialintelligenceact.eu/article/10/) 6. [AI RMF Core, NIST (2023)](https://airc.nist.gov/airmf-resources/airmf/5-sec-core/) 7. [Strategic AI Readiness: How To Move From Hype To Scalable Impact, Forrester (12 September 2025)](https://www.forrester.com/blogs/strategic-ai-readiness-how-to-move-from-hype-to-scalable-impact/) 8. [Add owners, Atlan docs](https://docs.atlan.com/product/capabilities/discovery/how-tos/add-owners) 9. [Add certificates, Atlan docs](https://docs.atlan.com/product/capabilities/discovery/how-tos/add-certificates) 10. [Manage custom metadata structures, Atlan docs](https://docs.atlan.com/product/capabilities/governance/custom-metadata/how-tos/manage-custom-metadata-structures) 11. [Automate data governance, Atlan docs](https://docs.atlan.com/product/capabilities/governance/stewardship/how-tos/automate-data-governance) 12. [Create governance workflows, Atlan docs](https://docs.atlan.com/product/capabilities/governance/stewardship/how-tos/create-governance-workflows) 13. [Metadata policy, Atlan docs](https://docs.atlan.com/product/capabilities/governance/access-control/metadata-policy) 14. [Use Atlan AI for documentation, Atlan docs](https://docs.atlan.com/product/capabilities/atlan-ai/how-tos/use-atlan-ai-for-documentation) 15. [Metadata enrichment FAQ, Atlan docs](https://docs.atlan.com/product/capabilities/governance/context-agents-studio/faq/metadata-enrichment) 16. [Design access, personas, and governance, Atlan docs](https://docs.atlan.com/product/capabilities/governance/access-control/best-practices/design-access-and-personas) 17. [What are data products, Atlan docs](https://docs.atlan.com/product/capabilities/data-products/concepts/what-are-data-products) 18. [Atlan MCP overview, Atlan docs](https://docs.atlan.com/product/capabilities/atlan-ai/how-tos/atlan-mcp-overview) 19. [Atlan MCP tools, Atlan docs](https://docs.atlan.com/product/capabilities/atlan-ai/references/mcp-tools) 20. [What is Context Engineering Studio, Atlan docs](https://docs.atlan.com/product/capabilities/governance/context-engineering-studio/concepts/what-is-context-engineering-studio) ## Next steps - [See how Atlan's context layer works in recorded demos](https://atlan.com/context-layer-demo/) - [Book a demo or talk to Atlan's sales team](https://atlan.com/forms/talk-to-sales-contact/)