> ## Content Index > Fetch the complete content index at: https://www.ratneshmishra.com/llms.txt > Use this file to discover other available public pages before exploring further. # The Enterprise Metadata Strategy: A Name in an Owner Field Is Not Accountability - URL: https://www.ratneshmishra.com/the-enterprise-metadata-strategy-a-name-in-an-owner-field-is-not-accountability/ - Published: 2026-09-20T21:18:32.000Z - Updated: 2026-09-20T23:23:03.000Z - Author: Ratnesh Kumar Mishra > “If an incident occurs at 2 a.m., the operating model must know exactly which producer team gets paged and which accountable seat owns the resulting risk.” --- For the CEO, CIO, CTO, CDO, and the executives who sit with them: risk, finance, compliance, and the lines of business that will be asked who was accountable. *Previously: metadata should be logically central and physically federated (Part 1), and AI plus a confidence band closes the coverage gap (Part 2). This week: the accountability layer — and why it is now an examinable control.* ![](https://storage.ghost.io/c/bf/09/bf092b96-7c12-4ea4-b204-9fb1e04f4342/content/images/2026/09/ChatGPT-Image-Sep-20--2026--04_17_01-PM.png) ### Why this is a C-suite issue, not a catalog issue A CEO does not need another coverage dashboard. A CEO needs to know whether, when a risk number cannot be reconstructed or an agent acts on the wrong table, a named seat still answers. That is safety and soundness. It is also franchise risk. A CIO and CTO own the systems that create, move, and change the data. If producer is missing from the model, every incident still lands on infrastructure — and every exam still asks who changed the transformation. A CDO owns the operating model that makes those seats real: four roles, inferred from how the estate already works, bound to deploy, access, certification, and incidents. Without that model, the catalog is a directory the board will be told is “green.” ### Regulators have already made this an accountability issue This is not theoretical. In October 2020 the OCC assessed a $400 million civil money penalty against Citibank. The public release named four deficiencies together: enterprise-wide risk management, compliance risk management, data governance, and internal controls. Inside the consent order is a phrase worth reading twice. > ***“An absence of clearly defined roles and responsibilities.”*** That sentence was not about a blank catalog column. It sat inside a broader failure of risk programs and Heightened Standards. That is exactly why it matters. Supervisors treat unclear roles as a safety-and-soundness defect, not as metadata hygiene. In July 2024 the OCC amended the 2020 order and assessed a further $75 million civil money penalty for violations of that order and for the lack of processes to monitor the impact of data-quality concerns on regulatory reporting. The Acting Comptroller called out persistent weaknesses “in particular with regard to data.” The same week, the Federal Reserve assessed $60.6 million against Citigroup under the related holding-company order. Across the related OCC and Federal Reserve actions, public penalties now exceed half a billion dollars. Most governance articles treat ownership as a field to populate before the audit. In banking it stopped being hygiene a decade ago. It is a control. It is examined. When remediation stalls, the second bill arrives with a press release. ### **Ten years of a binding standard** BCBS 239 was published in January 2013 and G-SIBs were expected to achieve compliance by January 2016\. Principle 1 requires clear accountability for risk data. The Basel Committee has told banks, repeatedly, to designate owners and take data-quality measures to the board. The Committee’s seventh progress report, published in November 2023, covered 31 G-SIBs. Average compliance across the principles moved from 3.14 in 2019 to 3.17 in 2022 on a four-point scale. Only two banks were fully compliant with all principles. No single principle was fully implemented by all banks. Assessments of the governance-related principles went backwards over that window. > **A decade of investment. Three hundredths of a point. That is not a vendor problem. Vendor problems get solved by the third RFP.** European supervision reached the same conclusion from another direction. The ECB’s Guide on effective risk data aggregation and risk reporting, published in May 2024, puts responsibility of the management body first. It tells significant institutions to select one or two members of the management body, in its management function, to exercise responsibility for implementing the data-governance framework — without discharging the body as a whole. The aggregated results of the 2025 SREP kept RDARR among the structural weaknesses in internal governance. About one in five internal-governance measures in that cycle concerned risk data aggregation and reporting, sitting alongside measures aimed at the management body itself. The ECB’s 2025 Annual Report on Supervisory Activities noted gradual progress once RDARR sat with senior management. The mechanism that moved the needle was not a new platform. It was naming someone who could be held. ## **One field, four accountabilities** Open almost any bank metadata repository and you will find a column called Owner. It holds one value. That single field is being asked four questions that have four different answers. ![](https://storage.ghost.io/c/bf/09/bf092b96-7c12-4ea4-b204-9fb1e04f4342/content/images/2026/09/ChatGPT-Image-Sep-20--2026--04_19_31-PM.png) | **Role** | **What it decides** | **Who the C-suite should recognize** | | ----------------- | ------------------------------------------------------------------ | --------------------------------------------------------- | | Accountable owner | Yes or no: risk, policy, certification, permitted use, retention | Domain executive or LOB data officer — a seat with budget | | Steward | Meaning: definitions, classification, quality rules, CDE inventory | Domain SME, senior enough to decide, not only to collect | | Producer | Change: schema, pipeline, deploy, runtime health | Engineering / platform — the CIO/CTO line | | Consumer | Use: confirms meaning, files the incident when the number is wrong | The desk that lives on the number | One human cannot be all four. Force it and you get a steward paged for an outage, a producer asked to certify a regulatory use, and a consumer listed as owner because they complained once. The producer role is the one the catalog almost never holds. Commit history, DAG configuration, and the CMDB already know who changes the asset. When an examiner asks who altered the transformation behind a risk feed, that is the name they want — and it is rarely the name in Owner. Consumers belong in the model because the person who queries a column every Tuesday notices when the number is wrong, and is the best reviewer of what it means. Consumers confirm meaning. They do not carry regulatory risk. > **Write the four fields. Resolve them separately. Display a single “talk to” only as a derived routing hint — never as the model.** ## Own by group, never by individual People change teams, take leave, and resign. An owner field containing a personal email has a short half-life. Nobody notices the decay until something breaks. The test is blunt: if an ownership record cannot route you to a durable organizational group when something matters, it is decorative. Every role should resolve to a group in the corporate directory. The producer should resolve to the on-call path the bank already trusts for production incidents; accountable owners, stewards, and consumers should resolve to equally durable governance and business-routing groups. ## What large U.S. banks have said in public — pattern, not a tour None of the following is an inside look. It is the public shape. Use it to recognize the pattern, not to copy a logo. ### **JPMorgan Chase — ownership bound to a product, not a label** Published materials describe a firmwide Chief Data and Analytics Office, lines of business owning the data closest to them, and data products curated by people who understand permissible use. Public architecture write-ups of that mesh bind data-product owners to entitlement — ownership as the mechanism that grants access, not as a string on a catalog card. The bank has also said the hard part was not the platform. It was agreeing what a data product is. ### **Bank of America — domain authority as an executive seat** Public roles describe domain data authorities accountable for strategy, governance, architecture alignment, and design oversight of a subdomain, including governed reporting. That is ownership as a seat with design rights. ### **Truist — policy, platform, and the team that builds them** In August 2022 Truist announced it had acquired Zaloni’s Arena platform and brought the product’s leadership and a cohort of product, engineering, and data professionals into the Enterprise Data Office. The public signal is not the brand of catalog. It is the willingness to treat metadata capability as something you staff, not something you license and hope the business fills in. ### **First Citizens — stewards in the business** In a published interview, the bank’s chief data and analytics officer named deepening and strengthening the role of data stewards across every business unit as a key priority, so that data is treated as a critical business asset. He made a second point that generalizes: an effective data leader connects the data strategy to the enterprise’s own drivers of profitability, growth, and customer satisfaction. That alignment is what sustains the investment across the years this work takes. ### **Wells Fargo — two public actions, two different lessons** In March 2026 the Federal Reserve terminated Wells Fargo’s 2018 enforcement action on governance oversight and compliance and operational-risk management, after nearly a decade of remediation and third-party reviews. That order was not a data-catalog order. Read it for the time scale. Governance that exists only on paper takes a decade to become examinable. Separately, the OCC’s 2024 formal agreement with Wells Fargo Bank requires a written data-integrity program for BSA/AML and OFAC systems: inventories of systems that hold relevant data, clear roles and responsibilities, data dictionaries and sourcing maps, lineage, a register of defects, testing, and oversight. Those requirements are scoped to financial-crimes systems. They still show what a supervisor means by owned metadata: every artifact has a seat, and the seat can be examined. ## **Four moves that change the outcome** **1\. Infer ownership, do not campaign for it** The campaign is always the same. Export the inventory. Ask each domain to claim assets. Put a human name on every row. Report percent-owned to the steering committee. It works for a quarter. Then people move, the name was never the right unit, claiming stays political, and nothing consumes the field. ![](https://storage.ghost.io/c/bf/09/bf092b96-7c12-4ea4-b204-9fb1e04f4342/content/images/2026/09/ChatGPT-Image-Sep-20--2026--04_04_51-PM.png) **A claim campaign asks people to volunteer. Inference watches who already acts.* | **Signal already in the estate** | **What it usually implies** | | ---------------------------------------------------- | ----------------------------------------------- | | Commits, CODEOWNERS, model authorship | Producer | | Pipeline or DAG owner, on-call roster | Producer, operational | | CMDB or application owner, cost center | Accountable seat, or the org that must name one | | Access-request approvers | Accountable owner or steward | | Query logs and warehouse consumption | Consumers — and the best reviewers of meaning | | Who answered the last data-help thread | Steward or expert consumer | | Who signed the last certification or CDE attestation | Accountable owner | | Who was paged the last time it broke | Producer, until proven otherwise | Propose the result as Suggested and let a human confirm. Confirming a pre-filled list finishes. Authoring a blank one does not. Inference is not gospel. A frequent querier is a consumer, not the risk owner. A last committer may have been fixing a typo. Rank signals. Require agreement across at least two sources before auto-filling even a Suggested producer. Escalate conflicts. Two domains committing to the same table is an architecture problem. > **When inference and a claim disagree, inference is the prior. The claim has to bring evidence.** **2\. Put ownership on the critical path** Metadata that nothing depends on will rot. Ownership is no different. • A deploy without a named producer fails the pipeline. • An access request without an accountable owner or steward cannot approve itself. • A certified or CDE asset that loses its accountable seat drops out of Verified. • An incident with no producer routes to the domain mailbox, not into silence. • Review items route to the top consumers plus the steward, not to a general queue. • Assets with no owner and no queries for ninety days enter deprecation: notify, then read-only, then archive. The day ownership is visible only inside the catalog portal, it is already rotting. The day it gates deploy, access, certification, and incident routing, people open the catalog because that is how they ship. Orphaned data is a cost line, a breach surface, and an unowned retention obligation ![](https://storage.ghost.io/c/bf/09/bf092b96-7c12-4ea4-b204-9fb1e04f4342/content/images/2026/09/ChatGPT-Image-Sep-20--2026--04_06_14-PM.png) > **3\. Make attestation the heartbeat** On a fixed cycle, a named accountable executive signs that the critical data elements in their domain are inventoried, classified, owned, and within quality tolerance — or declares the exceptions. Three things make attestation real: it is signed by a person with budget authority; the exceptions are enumerated rather than waved through; and the results flow into the operational-risk framework. A finding inside an RCSA is one the first line has to answer for. A finding that lives only in the catalog is a footnote. **4\. Publish domain scorecards where an executive will see them** Per-domain metrics on ownership coverage, classification completeness, CDE inventory currency, freshness-SLA attainment, and stale-asset count. Reviewed by the data council. Visible to the executive who owns the framework. A domain lead with twelve objectives attends to the one that appears on a dashboard their director reads. ## **The catalog is a product, not an inventory** An inventory is complete. A product is used. Asset counts do not predict whether a new analyst can find the right table in three minutes and know whether to trust it. • Search must match intent, not strings. • Trust signals belong on the result card: certification, freshness, quality, deprecation, who on your team queries it. • Embed the catalog in the SQL editor, the BI tool, and the notebook. • Curate in tiers. Holding 400,000 assets to one standard means holding none of them to it. ## **Why this gets harder in the next eighteen months** Every large institution is putting agents in front of enterprise data. When a person queries an unowned table and gets a wrong answer, a human in the loop may notice. When an agent retrieves from an unowned table, composes an answer, and acts, there is no such check. Model risk management has required named accountability for models since SR 11-7\. The data those models consume is now the weaker link. Institutions that cannot name an owner for a training set or a retrieval corpus will discover that gap during an examination rather than before one. > ***An agent acting on data with no owner is an automated decision with nobody accountable for its inputs.*** ## **What good looks like** | **Measure** | **12-month target** | | -------------------------------------------------------------- | ----------------------- | | Ownership coverage on tier 0–1 and CDE assets | \>95% | | Owner records resolving to a group, not a person | 100% | | Critical data elements with a named accountable executive seat | 100% | | Attestation cycle completed on schedule, exceptions enumerated | Every cycle | | Median time to reach a decision-maker on a data question | <1 business day | | Assets with no owner and no usage for 90 days | Declining every quarter | | Domains with a published governance scorecard | All in scope | | Data-governance findings flowing into RCSA | 100% | ### Three questions for the next executive forum • Take the three most business-critical data assets and page the owner. Not email — page. What happens next tells you whether you have ownership or a spreadsheet about ownership. • Which single named executive owns the risk-data framework? If the answer is a committee, you do not have an answer. • What breaks if the owner field is wrong? If nothing breaks, you have a directory. --- > Sources, all public: OCC Consent Order and Civil Money Penalty against Citibank, October 2020 (NR 2020-132 and related order), and the related Citigroup Form 8-K; OCC News Release 2024-76 and the concurrent Federal Reserve action, July 2024; OCC formal agreement with Wells Fargo Bank, N.A., September 2024, Article X (data integrity program for BSA/AML and OFAC systems); Basel Committee on Banking Supervision, Principles for Effective Risk Data Aggregation and Risk Reporting, and its seventh progress report, November 2023; ECB Guide on effective risk data aggregation and risk reporting, May 2024; ECB Aggregated Results of the 2025 SREP and Annual Report on Supervisory Activities 2025; Federal Reserve and Wells Fargo newsroom announcements of the termination of the 2018 consent order, March 2026; JPMorgan Chase published materials on its data-mesh and data-product architecture; Truist announcement of its acquisition of the Zaloni Arena platform, August 2022; published interview with the chief data and analytics officer of First Citizens Bank (CDO Magazine). > *This article draws only on publicly available information. It does not describe, and should not be read as describing, the practices of the author’s employer. Views are the author’s own.* *Ratnesh Kumar Mishra writes The Enterprise Metadata Strategy at ratneshmishra.com.*