Agentic AI consumes all three simultaneously. Govern only the data and metadata, and AI produces outputs that are technically accurate but contextually wrong.
The Evidence
The raw representation of business activity in the AI platform — built as data products with quality and access controls embedded as code.
Use it for
Focus areas
Goal
Trust
The Translator
The connective tissue that explains meaning and relationships as data hops from transactional to analytics layers for BI — and, in future, AI.
Use it for
Focus areas
Goal
Discoverability & Search
The Missing Ingredient
The situational nuance that turns information into insight — the "why" and "how" that AI needs to avoid being confidently wrong.
Use it for
Focus areas
Goal
AI Readiness
Metadata is descriptive; context is interpretive.
Metadata answers the what, where, when, who. Context answers the why and how — and it lives in people, so it's the first thing lost when they leave.
Governance in the Age of AI
Agentic AI consumes them simultaneously. It excels at processing metadata — structure and scale — but struggles with context, the meaning and intent, unless it is explicitly governed.
Give AI metadata without human context and it produces outputs that are technically accurate but contextually misleading. Deploying AI without governing context leads to hallucinations.
Because context is derived from human experience, culture, and business priorities, it cannot be fully automated. Treat context as a governable asset for explainability and trust.
Modernize governance into a source of acceleration rather than friction — so domains, data engineering, and platforms can sprint toward AI readiness.
Align with existing work behaviors and the domain organizational structure — leverage EDPM modernization for discovery, trust, and AI readiness.
Drive behavioral consistency across domains, platforms, and upstream business toward adoption of certified, compliant data products.
Enlist upstream stewards and SMEs to build the knowledge context for Agentic AI — resolving entity conflicts through the Council.
AI does not fix poor data — it amplifies it. Six characteristics are in scope under EDPM Track 1; four remain open gaps on the road to AI readiness.
Clear accountability
An explicitly recognized accountable party for how data is defined, produced, protected, and used.
Defined meaning
Shared, documented definitions so terms like “active customer” mean one thing across domains.
Verified quality
Accuracy measured with transparent thresholds and continuous monitoring — native Snowflake checks.
Transparency
Traceable origins and transformations from source to consumption, supporting impact analysis.
Risk-informed access
Row access policies and dynamic data masking aligned to sensitivity classifications.
Regulatory alignment
GDPR, PII, German Works Council, EU AI Act, DRDC, and data sovereignty embedded into the lifecycle.
Fairness oversight
Evaluating representativeness and fairness before and during training. A proposed Bias_Score tag.
Managed drift
Governing continuous change — comparing historic lineage snapshots before vs. after a change.
Right data, right use
Evaluating data against the model’s objective, error tolerance, and decision impact.
Always defensible
Ongoing, evidenced validation against standards — not a one-time slide. CETO Dashboard is step one.
Centralized policies and tools, executed by the domains closest to the data. The result: agility without losing consistency, trust, or compliance.
Embed governance into the data product lifecycle — not bolted on after deployment. The single highest-leverage commitment in the strategy.
Every product certified as accurate, traceable, securely accessed, and compliant for BI or AI.
The Council partners with the business to define the enterprise’s shared language.
Harmonize semantic layers across Cisco so AI delivers consistent answers everywhere.
Consolidate fragmented knowledge into a federated graph that gives AI deterministic context.
Data Governance Features
| Pillar | Focus | Stewardship role | Strategic goal |
|---|---|---|---|
| Metadata | Meaning & Lineage | Data Product Owner | Standardize language and map the data journey. |
| Data | Accuracy & Quality | Data Steward (upstream) | Ensure data is reliable, accessible, and fit for purpose. |
| Context | Interpretation & Intent | Domain Owner (AI Steward) | Capture the “why” to enable trusted decisions; resolve conflicts. |
The foundation for scalable, autonomous data management in the Agentic AI era. Each principle reinforces the others; partial adoption defeats the model.
Ownership distributed to the business domains closest to the data, managing the full lifecycle — ingestion, transformation, quality, delivery.
Eliminates central bottlenecksDomains treat data like a product: discoverable, understandable, trustworthy, addressable, accessible, secure, interoperable, and valuable.
8 essential characteristicsA central platform team provides domain-agnostic tools — catalog, governance automation, policy manager — readying the stack for Agentic AI.
Build vs. buy, standardizedA central council defines global standards; domains implement locally within guardrails, enforced computationally via platform automation.
Council guardrailsPrinciple 2 expansion · Data as a Product
A complete data product includes: dataset, metadata, code, contracts, tests, documentation, infrastructure.
Benefits of getting all four right: Improved agility · Risk mitigation · Enhanced trust · Increased participation
Organizational shift
Things that derail mesh adoption
