Data Governance & Quality Framework
Enterprise Data Management Guide

1. Why Data Governance Matters

Data governance is no longer a discretionary initiative for compliance-conscious organizations - it is a strategic imperative that directly determines an enterprise's ability to compete with AI, meet regulatory obligations, and extract reliable insight from an ever-expanding data estate. The organizations that treat data as a managed asset consistently outperform those that treat it as a byproduct of operational systems.

Gartner's ongoing research estimates that poor data quality costs organizations an average of $12.9 million per year. This figure encompasses direct costs (incorrect decisions, failed processes, manual remediation) and opportunity costs (delayed analytics projects, abandoned ML models, regulatory penalties). For enterprises operating across APAC markets - where regulatory fragmentation multiplies compliance complexity - the cost of ungoverned data compounds rapidly.

1.1 The Regulatory Imperative

The global regulatory landscape has shifted decisively toward accountability-based data protection. GDPR established the template, and APAC jurisdictions have followed with their own comprehensive frameworks: Singapore's PDPA (amended 2021 with mandatory breach notification), Thailand's PDPA (fully effective June 2022), Vietnam's Decree 13/2023/ND-CP on personal data protection, and sector-specific mandates from financial regulators (MAS in Singapore, Bank of Thailand, State Bank of Vietnam). Each framework demands that organizations demonstrate control over their data - knowledge of what personal data they hold, where it resides, how it flows, and who has access. Without formal governance, compliance becomes a perpetual firefight rather than a managed process.

1.2 AI and ML Data Requirements

The AI revolution has exposed a fundamental truth: machine learning models are only as reliable as the data they consume. Organizations investing millions in AI infrastructure frequently discover that their data is too fragmented, inconsistent, or poorly documented to support model training. Research from MIT Sloan and IBM consistently shows that data scientists spend 60-80% of their time on data preparation and quality remediation rather than actual modeling. A robust governance framework transforms this dynamic by ensuring data is discoverable, documented, quality-controlled, and lineage-tracked before it reaches the data science team.

1.3 Business Value of Governed Data

Beyond risk mitigation, data governance directly enables business value creation. Organizations with mature governance programs report:

• Faster time-to-insight: Data consumers find and trust data in hours rather than weeks, eliminating the "data scavenger hunt" that plagues ungoverned environments.
• Reduced redundancy: A governed data catalog eliminates shadow datasets and redundant ETL pipelines, reducing storage and compute costs by 20-35%.
• Improved decision quality: When business leaders trust their data, they make faster, more confident decisions - McKinsey research links data-driven decision-making to 23% higher revenue growth.
• M&A readiness: Due diligence for mergers and acquisitions increasingly scrutinizes data assets. Governed data with clear lineage and quality metrics commands a valuation premium.
• Customer experience: Master data management ensures a single, accurate view of the customer across all touchpoints - eliminating the duplicate mailings, mismatched records, and fragmented service histories that erode trust.

2. Governance Framework Design - DAMA DMBOK & Operating Models

Effective data governance requires a structured framework that defines the organizational model, roles, processes, and standards for managing data across the enterprise. The DAMA International Data Management Body of Knowledge (DMBOK2) provides the most widely adopted reference architecture, organizing data management into eleven knowledge areas with governance as the central coordinating function.

2.1 DAMA DMBOK Framework Overview

DAMA DMBOK2 defines data management as "the development, execution, and supervision of plans, policies, programs, and practices that deliver, control, protect, and enhance the value of data and information assets throughout their lifecycles." The framework organizes data management into eleven interrelated knowledge areas:

2.2 Data Governance Council

The governance council is the executive decision-making body for data management. Its composition, authority, and operating rhythm determine whether governance succeeds or becomes an ineffective committee. An effective council structure includes:

• Executive sponsor (CDO/CIO): Provides authority, budget, and escalation path. Without executive sponsorship, governance initiatives consistently fail within 12-18 months.
• Domain data owners: Senior business leaders (VP/Director level) accountable for data within their business domain - e.g., VP Sales for customer data, VP Finance for financial data, VP Supply Chain for product and inventory data.
• Chief Data Steward: Operational leader who coordinates the data stewardship network, manages the governance backlog, and reports progress to the council.
• Enterprise Architect: Ensures governance decisions align with the overall technology architecture and data platform strategy.
• Legal/Compliance representative: Ensures governance policies satisfy regulatory obligations across all operating jurisdictions.
• Data Engineering lead: Provides technical feasibility assessment for governance initiatives and implements approved changes to data pipelines and systems.

2.3 Data Governance Roles

Clearly defined roles with explicit accountabilities are the foundation of operational governance. The three core roles - data owner, data steward, and data custodian - form a hierarchy of accountability from strategic to operational to technical:

3. Data Quality Dimensions - Measurement & Monitoring

Data quality is not a binary state - it is a multi-dimensional measure of fitness for purpose. The six core dimensions, originally formalized by DAMA and refined through ISO 8000, provide a comprehensive framework for assessing, measuring, and monitoring the quality of any data asset. Each dimension requires specific measurement methods and different remediation strategies.

3.1 The Six Core Dimensions

3.2 Implementing Data Quality Rules

Quality rules should be defined collaboratively between data stewards (who understand business context) and data engineers (who implement technical checks). Rules are typically categorized into three tiers based on severity and action:

• Critical rules (blockers): Violations halt pipeline execution and trigger immediate alerts. Examples: primary key uniqueness violations, null values in mandatory fields, referential integrity breaks. These protect downstream systems from corrupted data.
• Warning rules (monitors): Violations are logged and flagged for investigation but do not block data flow. Examples: unusual distribution shifts, completeness below threshold, timeliness SLA breaches. These detect degradation trends before they become critical.
• Informational rules (profiling): Continuous statistical profiling that establishes baselines and detects anomalies. Examples: column cardinality changes, value distribution shifts, volume anomalies. These provide early warning of upstream system changes or data drift.

3.3 Data Quality Scoring

Aggregate quality scores provide a single-number summary of data fitness across dimensions. A weighted scoring model allows organizations to prioritize dimensions based on business impact. A typical scoring formula:

4. Data Catalog & Discovery - Metadata Management

A data catalog is the single most impactful investment an organization can make to accelerate data democratization. It serves as the enterprise's searchable inventory of data assets, combining technical metadata (schemas, data types, storage locations), business metadata (definitions, owners, sensitivity classifications), and operational metadata (lineage, quality scores, usage statistics) into a unified discovery interface.

4.1 Metadata Layers

Comprehensive metadata management addresses three distinct layers, each serving different stakeholders and use cases:

• Technical metadata: Schema definitions, column data types, table relationships, partition keys, storage format (Parquet, Delta, Iceberg), physical location (S3 bucket, database server), access credentials reference. Consumers: data engineers, DBAs, platform teams.
• Business metadata: Human-readable definitions, business glossary terms, data domain classification, data sensitivity level, data owner, regulatory tags (PII, PHI, financial), usage guidelines, quality SLAs. Consumers: data analysts, business users, compliance officers.
• Operational metadata: Data lineage (upstream sources and downstream consumers), pipeline execution history, freshness timestamps, quality check results, access logs, query statistics, cost attribution. Consumers: data engineers, governance teams, FinOps.

4.2 Business Glossary

The business glossary is arguably the most valuable component of a data catalog for non-technical stakeholders. It provides an authoritative, organization-wide dictionary of business terms with precise definitions, approved by data owners. Without a glossary, the same term frequently means different things across departments - "active customer" might mean "purchased in last 12 months" to Sales but "has a valid contract" to Finance, leading to conflicting reports and eroded trust in data.

A well-maintained business glossary includes:

• Term name: The canonical business term (e.g., "Annual Recurring Revenue").
• Definition: Precise, unambiguous definition approved by the data owner ("Sum of annualized contract values for all active subscriptions as of the measurement date, excluding one-time fees and professional services").
• Synonyms and abbreviations: Alternative names used across the organization (ARR, annual recurring, subscription revenue).
• Data owner: The business leader accountable for this term's definition and usage.
• Related terms: Cross-references to related glossary entries (Monthly Recurring Revenue, Net Revenue Retention, Customer Lifetime Value).
• Linked data assets: Database columns, reports, and dashboards where this metric is implemented.
• Calculation formula: The precise computation logic, including edge cases and exclusions.

4.3 Data Lineage Tracking

Data lineage traces the complete journey of data from source systems through transformations to consumption endpoints. It answers the questions: "Where did this data come from?", "What transformations were applied?", and "What downstream systems or reports will break if this data changes?" Lineage is essential for impact analysis, regulatory compliance (GDPR right to erasure requires knowing everywhere personal data flows), and debugging data quality issues to their root cause.

Modern lineage tracking approaches include:

• SQL parsing: Tools like Atlan, Collibra, and OpenLineage parse SQL queries from dbt, Airflow, Spark, and data warehouse query logs to automatically construct column-level lineage graphs. This is the most common approach and requires no instrumentation of existing pipelines.
• API instrumentation: OpenLineage (Linux Foundation project) provides a standardized API for emitting lineage events from any data processing framework. Supported by Airflow, Spark, dbt, Flink, and Great Expectations. Enables real-time lineage capture at execution time.
• Metadata extraction: Crawlers connect to databases, data lakes, BI tools, and orchestrators to extract schema information and infer lineage from naming conventions, foreign key relationships, and ETL job configurations.

4.4 Automated Data Cataloging

Manual cataloging does not scale. An enterprise with 50+ source systems, thousands of tables, and millions of columns cannot rely on manual documentation. Modern data catalogs employ automated discovery including:

• Schema crawlers: Automatically discover and catalog database schemas, table structures, and column metadata on a scheduled basis. Support for JDBC, ODBC, Hive Metastore, AWS Glue Catalog, and cloud-native APIs.
• PII detection: ML-based classifiers that scan column names and sample data values to automatically tag columns containing personal information (names, emails, phone numbers, national IDs, credit card numbers). Critical for privacy compliance.
• Usage analytics: Track which datasets are queried most frequently, by whom, and for what purpose. Identifies high-value datasets that deserve investment in documentation and quality, as well as orphaned datasets that can be retired.
• Social curation: Allow data consumers to rate, comment on, and tag datasets. The catalog becomes a living knowledge base where tribal knowledge is captured alongside technical metadata.

5. Master Data Management - Golden Records & Entity Resolution

Master Data Management (MDM) is the discipline of creating and maintaining a single, authoritative version of critical business entities - customers, products, suppliers, employees, locations - that is consistent across all enterprise systems. The "golden record" represents the best-known, most complete, and most accurate representation of an entity, assembled from multiple source systems through matching, merging, and survivorship rules.

5.1 MDM Architecture Styles

MDM implementations follow one of four architectural patterns, each with distinct trade-offs in terms of complexity, data latency, and organizational impact:

5.2 Entity Resolution

Entity resolution (also called record linkage or deduplication) is the process of determining whether two records in one or more datasets refer to the same real-world entity. This is a core MDM capability that addresses the uniqueness dimension of data quality. Modern entity resolution combines multiple techniques:

• Deterministic matching: Exact match on high-confidence identifiers (tax ID, email, phone number). Fast and precise but misses records with data entry variations.
• Probabilistic matching: Weighted scoring across multiple attributes using algorithms like Fellegi-Sunter. Handles variations, abbreviations, and transpositions. Requires careful threshold tuning to balance precision and recall.
• ML-based matching: Supervised learning models trained on labeled match/non-match pairs. Outperforms probabilistic methods on complex datasets but requires training data. Libraries like Splink, Dedupe.io, and Zingg provide production-ready implementations.
• Graph-based resolution: Constructs a network of relationships between records and uses community detection algorithms to identify clusters of records that likely represent the same entity. Effective for resolving complex hierarchies (parent/subsidiary companies).

5.3 Survivorship Rules

When multiple records are matched to the same entity, survivorship rules determine which attribute values are selected for the golden record. Common strategies include:

• Source system priority: Values from the most authoritative source for each attribute win (e.g., legal name from ERP, email from CRM, billing address from payment system).
• Most recent: The most recently updated value wins, assuming newer data is more accurate.
• Most complete: The record with the most populated fields contributes priority values.
• Frequency-based: When the same value appears across multiple sources, it is preferred over outlier values.
• Manual curation: Data stewards review and manually select golden values for high-value or ambiguous entities.

6. Data Privacy & Compliance - GDPR, PDPA & Vietnam Law

Data governance and privacy compliance are inseparable. A governance framework that does not account for regulatory requirements is incomplete, while privacy compliance without governance infrastructure is unsustainably expensive. For APAC enterprises operating across multiple jurisdictions, the compliance landscape is particularly complex, with each market imposing distinct requirements for consent, data localization, breach notification, and cross-border transfer.

6.1 Regulatory Landscape Comparison

6.2 Data Classification Framework

Data classification is the governance mechanism that assigns sensitivity labels to data assets, driving access control, encryption, retention, and handling policies. A four-tier classification model is standard for enterprise governance:

• Public: Data intended for public consumption. No access restrictions required. Examples: published marketing content, public financial filings, product specifications.
• Internal: Data available to all employees but not for external disclosure. Standard access controls. Examples: organizational charts, internal policies, aggregate business metrics.
• Confidential: Data restricted to authorized personnel with a business need. Requires encryption at rest and in transit, access logging, and DLP monitoring. Examples: customer PII, employee records, financial data, contracts, strategic plans.
• Restricted: Highest sensitivity data requiring the strictest controls. Multi-factor authentication, row-level security, data masking for non-production environments, and comprehensive audit trails. Examples: payment card data (PCI DSS), health records (PHI), government secrets, cryptographic keys, authentication credentials.

6.3 Consent Management

Multi-jurisdictional operations require a consent management system that tracks individual consent preferences across all applicable regulations and data processing purposes. Key capabilities include:

• Purpose-specific consent tracking: Record consent for each processing purpose (marketing, analytics, service delivery, third-party sharing) independently.
• Jurisdiction awareness: Apply the correct legal basis (consent, legitimate interest, contractual necessity) based on the data subject's jurisdiction.
• Consent lifecycle management: Capture consent acquisition, store proof of consent, process withdrawal requests, and propagate preference changes to all downstream systems.
• API-driven enforcement: Expose consent status via API so that downstream systems (CRM, marketing automation, analytics) can check consent before processing.

7. Data Quality Tools & Technologies

The data quality and governance tooling landscape has matured significantly, with options spanning open-source frameworks for engineering-led organizations through enterprise platforms for large-scale governance programs. The right tool selection depends on organizational maturity, scale, existing data stack, and whether governance is driven primarily by engineering teams or business-side governance functions.

7.1 Data Quality Frameworks

7.2 Data Catalog & Governance Platforms

7.3 Integration Architecture

A production-grade data governance stack integrates quality tools, catalogs, and orchestrators into a coherent pipeline. The following architecture represents a common pattern for modern data stack environments:

8. Implementation Roadmap - From Assessment to Operating Model

Data governance programs fail most commonly not from lack of tools or frameworks, but from attempting too much too soon, failing to secure executive sponsorship, or neglecting the organizational change management required to embed governance into daily operations. The following phased roadmap is based on our experience implementing governance programs across APAC enterprises.

Phase 1: Assessment & Foundation (Months 1-3)

1. Governance maturity assessment: Evaluate current state across the DAMA DMBOK knowledge areas using a standardized maturity model (Stanford, CMMI DMM, or EDM Council DCAM). This establishes a baseline, identifies critical gaps, and provides an objective measure for tracking progress.
2. Stakeholder interviews: Conduct structured interviews with 15-25 stakeholders across business domains, IT, compliance, and data science to understand pain points, priorities, and political dynamics. The governance program must solve problems that stakeholders actually care about.
3. Critical data element identification: Identify the top 20-30 data elements that drive the most business value and/or regulatory risk. These become the initial scope of the governance program.
4. Executive sponsorship: Secure formal sponsorship from CDO/CIO with defined authority, budget commitment (typically 0.5-2% of total data/IT spend), and a visible mandate communicated to the organization.
5. Governance charter: Draft and approve a governance charter defining mission, scope, authority, organizational structure, and decision rights. This is the constitutional document for the governance program.

Phase 2: Quick Wins & Core Processes (Months 4-6)

1. Data quality profiling: Profile critical data elements across source systems to establish baseline quality metrics. Use Great Expectations, dbt tests, or Soda Core for automated profiling. Document current quality scores for each dimension.
2. Business glossary (initial): Define and publish glossary entries for the top 50-100 business terms covering critical data elements. Ensure definitions are approved by data owners and accessible to all data consumers.
3. Data ownership assignment: Formally assign data owners and data stewards for each critical data domain. Publish the responsibility matrix (RACI) and secure written acknowledgment from each role-holder.
4. First governance council meeting: Convene the governance council with a prepared agenda: review maturity assessment results, approve governance charter, endorse initial policies, and set quarterly objectives.
5. Quick win delivery: Identify and resolve 3-5 visible data quality issues that have been causing business pain. Nothing builds momentum for a governance program like demonstrating tangible value early.

Phase 3: Platform & Scale (Months 7-12)

1. Data catalog deployment: Select and deploy a data catalog platform (Atlan, Collibra, Alation, or OpenMetadata). Configure connectors to critical data sources. Seed the catalog with metadata from Phase 2 profiling and glossary work.
2. Automated quality monitoring: Implement automated quality checks for critical data elements in production pipelines. Configure alerting thresholds and incident response procedures.
3. Data lineage implementation: Deploy lineage tracking for critical data flows. Integrate with dbt, Airflow, and the data catalog to provide end-to-end visibility from source to report.
4. Policy formalization: Codify data governance policies covering: data classification, access request procedures, data quality standards, retention and archival, cross-border transfer, incident response, and acceptable use.
5. Training program: Develop and deliver governance training for data owners, stewards, engineers, and consumers. Include role-specific curricula and certification paths.

Phase 4: Optimization & Advanced Capabilities (Months 13-18)

1. Expand domain coverage: Extend governance to secondary data domains beyond the initial critical data elements. Onboard additional data stewards and expand the business glossary.
2. MDM implementation: If justified by business requirements, implement master data management for the highest-value entity domains (typically customer and product).
3. Self-service governance: Evolve toward a model where data consumers actively participate in governance through catalog curation, quality issue reporting, and glossary contribution.
4. Governance metrics dashboard: Build a comprehensive governance dashboard tracking maturity scores, quality trends, catalog usage, issue resolution metrics, and business impact KPIs.
5. Continuous improvement: Conduct second maturity assessment to measure progress from Phase 1 baseline. Adjust strategy based on findings and evolving business priorities.

9. Data Mesh & Decentralized Governance

Data Mesh, proposed by Zhamak Dehghani in 2019 and refined through her 2022 book, represents the most significant architectural paradigm shift in data management since the data warehouse. It challenges the centralized data team model that has dominated enterprise data management for two decades, proposing instead a decentralized, domain-oriented approach where data is treated as a product and governed through federated computational policies.

9.1 Four Principles of Data Mesh

• Domain-oriented ownership: Data ownership and responsibility shifts from a centralized data team to the business domains that produce the data. The sales domain owns and publishes sales data products; the supply chain domain owns inventory and logistics data products. Each domain has its own data engineers, stewards, and product owners.
• Data as a product: Data is treated with the same rigor as customer-facing software products. Each data product has an owner, SLAs (quality, freshness, availability), documentation, discoverability through a catalog, and a well-defined interface (API/schema). Data products are designed for consumers, not just emitted as byproducts of operational systems.
• Self-serve data platform: A centralized platform team provides the infrastructure, tooling, and abstractions that enable domain teams to create, publish, and consume data products without requiring deep infrastructure expertise. This includes provisioning, security, quality testing frameworks, catalog registration, and cost management - all as self-service capabilities.
• Federated computational governance: Governance policies are defined centrally but enforced computationally (through automated checks, platform guardrails, and policy-as-code) rather than through manual review processes. The central governance team sets interoperability standards, quality thresholds, and compliance requirements; domain teams implement them using platform-provided tools.

9.2 Federated Governance in Practice

Federated governance balances central standardization with domain autonomy. The central governance team is responsible for:

• Interoperability standards: Schema conventions (naming, data types), event formats, API contracts, and shared reference data (country codes, currency codes, industry classifications) that enable data products from different domains to be combined without translation.
• Global policies: Data classification framework, privacy compliance requirements, retention standards, security baselines, and audit requirements that apply uniformly across all domains.
• Quality baselines: Minimum quality thresholds that every data product must meet before publication (e.g., primary key uniqueness, schema documentation, freshness SLA, PII tagging).
• Platform guardrails: Automated enforcement of policies through CI/CD pipelines, infrastructure provisioning, and platform APIs. A data product that fails quality checks or lacks required documentation cannot be published to the catalog.

Domain teams retain autonomy over:

• Domain-specific quality rules: Business rules that are meaningful within the domain context (e.g., "order total must equal sum of line items" is a Sales domain rule, not a global standard).
• Data modeling decisions: How data is structured within the domain, as long as published interfaces conform to interoperability standards.
• Technology choices: Within the platform-provided toolkit, domains choose specific tools and patterns that best fit their requirements.
• Prioritization: Domains determine which data products to develop based on consumer demand and domain strategy.

9.3 When to Adopt Data Mesh

Data Mesh is not universally appropriate. It is most effective for organizations that meet specific criteria:

• Strong fit: Large organizations (500+ employees in data-producing roles) with multiple distinct business domains, mature engineering culture, and a centralized data team that has become a bottleneck.
• Moderate fit: Mid-size organizations with 3-5 data-producing domains and a desire to scale data capabilities without proportionally scaling the central data team.
• Poor fit: Small organizations (under 100 employees), early-stage data programs without basic governance infrastructure, or organizations where a single domain dominates data production. For these, a centralized or hub-and-spoke model is more pragmatic.

10. Measuring Success - Scorecards, Maturity & Business Impact

Governance programs that cannot demonstrate measurable value are perpetually at risk of defunding. Robust measurement requires a balanced set of metrics spanning operational effectiveness, data quality trends, and business impact - connecting governance activities to outcomes that executives care about.

10.1 Data Quality Scorecards

Quality scorecards provide an at-a-glance view of data fitness across critical data elements and dimensions. An effective scorecard structure includes:

• Domain-level scores: Aggregate quality scores for each business domain (Customer: 94%, Product: 89%, Financial: 97%, Supply Chain: 82%). Enables comparison and prioritization across domains.
• Dimension-level breakdown: Per-domain scores decomposed by quality dimension (Accuracy, Completeness, Consistency, Timeliness, Validity, Uniqueness). Identifies which aspects of quality need the most attention.
• Trend analysis: Month-over-month quality trends showing improvement trajectory. A score that has improved from 78% to 91% over six months demonstrates governance value even if it has not yet reached the 95% target.
• Critical data element drill-down: Individual quality scores for each critical data element, with rule-level detail showing which specific checks are passing and failing.
• SLA compliance: Percentage of data products meeting their published quality, freshness, and availability SLAs.

10.2 Governance Maturity Models

Maturity models provide a structured framework for assessing governance capability and tracking improvement over time. The most widely used models include:

10.3 Business Impact Metrics

The most compelling governance metrics connect data quality improvements to business outcomes. Track these metrics to demonstrate ROI to executive stakeholders:

• Decision latency reduction: Time from question to data-backed answer. Governed environments with catalogs and documented datasets typically reduce this from weeks to hours.
• Regulatory compliance cost: Time and resources spent on regulatory audit preparation and response. Mature governance reduces compliance preparation effort by 40-60%.
• Data incident frequency: Number of data quality incidents that impact business operations or reporting. Track severity, root cause, time to detection, and time to resolution.
• AI/ML model performance: Model accuracy and drift metrics correlated with input data quality scores. Demonstrates the direct link between governance investment and AI capability.
• Pipeline efficiency: Reduction in failed pipeline runs, data reprocessing, and manual data cleansing effort. Quantify in engineering hours saved per month.
• Customer data accuracy: Percentage of customer communications (invoices, marketing, support) sent with correct information. Directly impacts customer satisfaction and revenue leakage.
• Catalog adoption: Number of active catalog users, searches per month, datasets bookmarked, and glossary terms viewed. Leading indicator of governance program health.

11. Frequently Asked Questions

What is data governance and why does it matter for enterprise organizations?

Data governance is the framework of policies, processes, roles, and standards that ensures data is managed as a strategic enterprise asset. It matters because organizations with mature data governance reduce data-related errors by 60-80%, achieve 40% faster regulatory compliance, and unlock significantly higher ROI from AI/ML initiatives. Gartner estimates that poor data quality costs organizations an average of $12.9 million per year. For APAC enterprises operating across multiple regulatory jurisdictions, governance provides the structural foundation for consistent compliance without duplicating effort in each market.

What is the difference between a data steward, data owner, and data custodian?

A data owner is a senior business leader (typically VP or Director level) who is accountable for a data domain. They define data policies, approve access requests, set quality thresholds, and resolve cross-domain data conflicts. A data steward is a subject-matter expert who implements governance policies on a day-to-day basis, investigates and resolves data quality issues, maintains business glossary definitions, and serves as the bridge between business and technical teams. A data custodian is an IT professional responsible for the technical infrastructure: database administration, security control implementation, backup and recovery procedures, encryption, and physical storage management. All three roles must be filled and coordinated for governance to function.

What are the six core data quality dimensions?

The six core dimensions are: Accuracy (data correctly represents the real-world entity), Completeness (all required data values are present), Consistency (data does not contradict itself across systems), Timeliness (data is available when needed and reflects the current state), Validity (data conforms to defined formats, ranges, and business rules), and Uniqueness (each entity is represented only once without unwanted duplicates). Each dimension requires different measurement methods and remediation approaches. Organizations should weight these dimensions based on their specific business context - a financial institution will weight accuracy and consistency heavily, while a marketing team may prioritize completeness and timeliness.

How does Data Mesh differ from traditional centralized data governance?

Traditional centralized governance places a single data team in control of all data assets, policies, and quality. This works for smaller organizations but creates bottlenecks as data complexity scales. Data Mesh, proposed by Zhamak Dehghani, shifts to domain-oriented ownership where each business domain owns, produces, and serves its data as a product. Governance becomes federated: a central team defines interoperability standards, compliance policies, and quality baselines, while domain teams implement governance within those guardrails using self-serve platform tools. The key difference is that governance moves from gatekeeping (central team approves everything) to guardrailing (central team sets standards, automated enforcement ensures compliance, domains operate autonomously within bounds).

Which data governance tools are best suited for APAC enterprise deployments?

For large enterprises with formal governance programs, Collibra and Alation are the leading commercial platforms with strong APAC presence and local support. For cloud-native organizations using the modern data stack (Snowflake, dbt, Fivetran), Atlan offers a modern metadata platform with excellent integration and a more accessible price point. For open-source deployments, Apache Atlas (Hadoop ecosystem), OpenMetadata, and DataHub (LinkedIn) provide robust metadata management. Data quality specifically is well-served by Great Expectations (open-source), dbt tests (built into transformation layer), Monte Carlo (ML-driven observability), and Soda Core (declarative quality checks).

What compliance frameworks apply to data governance in Southeast Asia?

Key frameworks include: Singapore PDPA (Personal Data Protection Act) with mandatory breach notification, DPO requirements, and the 2021 amendment adding data portability; Thailand PDPA (fully effective June 2022) closely modeled on GDPR with explicit consent requirements; Vietnam's Decree 13/2023/ND-CP on personal data protection with significant data localization requirements and cross-border transfer impact assessments; and GDPR for any organization processing EU citizen data. Industry-specific regulations add additional layers: MAS TRM and MAS Technology Risk Management Guidelines for financial services in Singapore, Bank of Thailand IT risk management guidelines, and Vietnam's Cybersecurity Law (2018) with broad data localization provisions. A governance framework must map controls to all applicable regulations for each operating jurisdiction.