# Data governance: roles, rules and how it differs from data strategy and AI governance

Author: Marin Niehues, Niehues Consulting · Published: 2026-06-08 · Updated: 2026-11-16 · AI Law, Risk & Governance
Source: https://www.niehuesconsulting.com/blog/data-governance/

Data governance is the set of decision rights and roles that determines who decides about which data and what quality that data must have. Data strategy sets the goal, while AI governance regulates the use of AI systems. This article separates the terms, describes the roles and sets out a 90-day build. This article provides an overview and does not constitute legal advice.

AI has given the topic new urgency. According to official German statistics, around 26 percent of companies with ten or more employees used AI in 2025, rising to 57 percent among those with 250 or more. Every one of these applications works with company data. They repeatedly fail on unclear accountability and conflicting definitions, as well as on access that nobody approved. Data governance addresses exactly these causes.

**Data governance is a leadership task with clear decision rights. It takes effect once every business-critical data domain has a named accountable person from the business and quality is measured against agreed metrics.**

## What is data governance?

Data governance is the exercise of authority and control over the management of data, shared between business functions and IT. That is how DAMA-DMBOK, the reference work of the professional association DAMA International, defines it. The definition contains an important distinction: governance directs data management and does not replace it. Data management runs databases and interfaces. Governance sets the rules for how that happens and who decides in a conflict.

Functioning data governance consists of five elements:

- **Decision rights:** who sets definitions, who approves access, who decides when goals conflict?
- **Roles:** named people for each data domain, with a time budget and a mandate.
- **Policies and standards:** classification, quality requirements, access, retention.
- **Procedures:** approval of new data uses, remediation of quality issues, escalation.
- **Metrics:** measured quality per domain and the time it takes to approve a data use.

If any of the five is missing, governance remains a document. Decision rights are missing most often: policies exist, but nobody has the authority to settle a dispute between sales and finance over the definition of an "active customer".

## What is the difference between data strategy, data governance and AI governance?

Data strategy defines the business value a company wants to create with data. Data governance regulates the reliable handling of that data. AI governance regulates the development and use of AI systems. The three build on each other in that order and overlap at clearly identifiable points.

**How the three terms differ**
Data strategy: goals and priorities. It sets out which data is needed for which business outcomes and which capabilities the company builds for this. It covers several years and is reviewed annually.

Data governance: rules and accountability for all data, with or without AI. It applies to finance reports just as much as to a model's training data, and it runs permanently.

AI governance: rules for AI systems across their life cycle. It covers risk classification and approvals, human oversight and monitoring in operation, including for models trained on third-party data.

The overlap lies in the data that AI systems use. For high-risk systems, Article 10 of the EU AI Act explicitly requires data governance practices for training, validation and testing data, covering among other things data origin and examination for bias. AI governance builds on data governance at this point and draws on its roles. A company without data governance writes separate data rules for each AI application, and they contradict each other after a few applications. How the AI-side committees and policies are set up is covered in [building AI governance](/blog/ai-governance-framework/). Goals and priorities on the data side are covered in [data strategy as a prerequisite for AI](/blog/data-strategy-as-a-prerequisite-for-ai/).

## Which roles does data governance need?

Data governance needs four roles: the data owner, accountable for a data domain; the data steward, who maintains it day to day; the technical custodian in IT; and a small board for cross-functional decisions. The roles are tasks within existing positions. New positions usually arise only in coordination.

- **Data owner:** a business leader, for example the head of sales for customer data. The owner decides on definitions and quality targets, approves access and answers for the data to the executive team.
- **Data steward:** the subject-matter contact for a domain. The steward maintains definitions in the data catalog, monitors quality metrics and resolves issues with the people who capture the data. Technical data stewards in IT support this for interfaces and data models.
- **Data custodian:** the IT team that implements storage and access technically without deciding on content.
- **Data governance board:** the data owners of the most important domains, chaired by a member of the executive team or a chief data officer. It resolves conflicts between domains and adopts the policies.

As a planning assumption, allow one fifth of a data steward's working time, more for domains with many source systems. Without this time in the steward's objectives, the role exists only on paper. The data protection officer remains separate. The DPO advises and monitors under the GDPR, while decisions about the data rest with the data owner.

## Rules: what a data governance policy sets out

An effective policy covers six topics in a few pages and refers to domain standards for detail:

- **Classification:** protection levels from public to strictly confidential, with a rule on which levels may enter which AI tools.
- **Definitions:** a shared glossary for core terms such as customer and revenue, maintained in the data catalog.
- **Quality:** measurable requirements per domain. ISO/IEC 25012 describes 15 data quality characteristics. To start, completeness and accuracy are sufficient, supplemented by currency and consistency.
- **Access:** who may see which data, who approves access and how quickly.
- **Lineage:** traceable data flows from source to report or model.
- **Retention and deletion:** periods per data type, aligned with the record of processing activities under the GDPR.

The need for action on quality is well documented. In a measurement across 75 companies, managers assessed the last 100 data records of their unit. On average, 47 percent of newly created records contained at least one error affecting work, and only 3 percent of companies reached an acceptable score. Banking supervisors drew conclusions early: the Basel Committee on Banking Supervision sets out 14 principles for risk data aggregation, including requirements on governance and accuracy. How quality is measured for AI is covered in [data quality and AI success](/blog/data-quality-for-ai-success/).

**Evidence on data governance**
47 % | Share of newly created records with at least one work-affecting error, average across 75 companies
3 % | Share of companies in the same measurement with acceptable data quality
15 | Data quality characteristics under ISO/IEC 25012
14 | Basel Committee principles for risk data aggregation and risk reporting

## What is data mesh, and when does it fit?

Data mesh is an operating model in which business domains own and provide their data as products, on a shared platform and under shared rules enforced automatically. The concept was developed by Zhamak Dehghani, who described it from 2019 onwards. It rests on four principles:

- domain-oriented ownership of data
- data as a product with quality commitments to its users
- a self-serve data platform
- federated governance whose rules are enforced technically

Data mesh is therefore a form of data governance. Rules are agreed centrally and implemented locally. Clear criteria govern the choice of model. A company with fewer than four business units that run their own data teams is better served by central governance with clear data owners. From four such units and an established data platform, the central team becomes a bottleneck. Shifting accountability to the domains then speeds up delivery.

**Operating model for data governance**
Criterion | Central data governance | Data mesh | With Niehues Consulting
Suited to | Companies with few data domains | Several units with their own data teams and a mature platform | Model chosen by number of domains and maturity, transition planned
Accountability | Central data team with data owners | Domain teams for their data products | Data owner named for each domain, with a time budget
Typical risk | Bottleneck in the central team | Inconsistent definitions without strong rules | Roles and decision paths settled before technology
First measurable benefit | Consistent core metrics in reporting | Faster delivery of new data products | Baseline and target per domain agreed before the start

## How long does it take to build data governance?

Working data governance for the first two or three data domains is in place after around 90 days; extending it across the company takes one to two years. Sequence is decisive: start with the domains that the next AI application or the most important reporting depends on.

**A 90-day build**
Days 1 to 30: select domains, name data owners and stewards, measure baseline data quality.

Days 31 to 60: set core definitions and classification, set up the access approval procedure, establish the board.

Days 61 to 90: report quality metrics on a fixed cadence, resolve a first issue through the procedure, bring the policy into force.

A simple calculation helps with cost. Internal effort arises mainly with the stewards. Three domains with one steward each at 20 percent and 220 working days a year amount to around 130 person-days annually. Add licenses for a data catalog and, where needed, external support. External consulting in the German market is typically billed at around 1,000 to 2,500 euros per consultant day. Set against this is the cost that poor data causes today, from rework to AI applications stuck in pilot. How this fits into overall governance is described in the article [AI management: what executive leadership must steer](/blog/ai-management-what-executives-must-steer/).

## Niehues Consulting as your partner for data governance

[Niehues Consulting](/) builds data governance from a delivery perspective: starting with the domains the next AI application needs, and with roles that have time for the work. Marin Niehues is a management consultant. He builds organizations that deliver and brings AI into operations, in engagements ranging from VC-funded scale-ups to DAX 40 corporations. He has worked in 20+ countries on six continents, with the DACH region as his home base.

The approach follows the Niehues Delivery Model with its phases Understand, Build and Embed. Three of its check questions matter most for data governance. The first is: who decides, who delivers, who uses? It leads directly to data owner and steward. The second asks for the baseline and the measure of success, which requires a quality measurement before the first decision. The third asks which roles and decision paths operation requires. It anchors governance permanently in the line organization. The starting point is a free 30-minute initial consultation via [our contact page](/#kontakt).

**Four decisions for the executive team**
The executive team names one member accountable for data governance and a data owner from the business for every business-critical domain. It gives data stewards a fixed time budget. It selects the first domains according to the needs of the next AI application. It has quality measured against metrics and reported quarterly.

**Sources**
- Federal Statistical Office of Germany (Destatis): Use of information and communication technology in enterprises, 2025
- Nagle, Redman, Sammon: Only 3% of Companies’ Data Meets Basic Quality Standards, Harvard Business Review, 2017
- ISO/IEC 25012: Software engineering, Software product Quality Requirements and Evaluation (SQuaRE), Data quality model, 2008
- Basel Committee on Banking Supervision: Principles for effective risk data aggregation and risk reporting (BCBS 239), 2013
- Bundesverband Deutscher Unternehmensberatungen (BDU): Fees in management consulting, fee study 2022
- Regulation (EU) 2024/1689 laying down harmonized rules on artificial intelligence (AI Act), Article 10, 2024

## Frequently asked questions

### What is a data steward?

A data steward is the subject-matter contact for a data domain who maintains definitions and monitors data quality. The steward acts on behalf of the data owner, who takes the decisions. The role is usually part of an existing position in the business.

### What is the difference between a data owner and a data steward?

The data owner decides and is accountable; the data steward maintains and monitors. The owner is a leader with a mandate who sets definitions and quality targets. The steward puts those decisions into practice and reports deviations.

### What is the difference between data governance and data management?

Data governance sets rules and decision rights; data management carries them out operationally. Data management covers databases and integration. Governance determines the standards that apply and who decides in a conflict.

### Who is responsible for data governance?

Overall accountability sits with a member of the executive team or a chief data officer. Data owners in the business are accountable for individual data domains. IT implements the rules technically and does not decide on content.

### What is the EU Data Governance Act?

The Data Governance Act is an EU regulation on the re-use of protected public-sector data, on data intermediation services and on data altruism. It therefore governs data sharing between organizations. Despite its name, it does not prescribe a company's internal data governance.

### What is a data governance framework?

A data governance framework is the documented combination of decision rights, roles, policies, procedures and metrics that a company uses to govern its data. Reference works such as DAMA-DMBOK describe the components. A company adapts the framework to the number of its domains and its regulatory requirements.

### Which standards apply to data governance?

The main references are DAMA-DMBOK for terminology and knowledge areas, ISO/IEC 38505-1 for the governance of data by top management and ISO/IEC 25012 for data quality. Banks also follow the BCBS 239 principles. There is no certification requirement for data governance.

### Which tools does data governance need?

The central tool is a data catalog that shows definitions and owners together with data lineage in one place. Tools for measuring data quality and managing access complement it. Tool selection follows the roles and rules.
