Beyond Gold: Building an AI-ready data foundation for the agentic era

10-minute read

For the past decade, enterprise data strategies have been primarily focused on enabling analytics modernization. The goal was familiar: consolidate data from siloed source systems, clean and curate it, and surface it in dashboards and reports that business teams could trust. The traditional medallion architecture (Bronze, Silver, Gold) gave business users a clean, proven framework for that journey. It delivered real value, and it remains the right foundation.

But AI doesn’t consume data the way a dashboard does.

A dashboard presents information. A business intelligence report answers a specific question, scoped and authored by a human analyst who supplies the interpretive context. An AI system, by contrast, reasons across domains, traverses relationships between business entities, orchestrates automated workflows, and generates outputs that downstream systems act on, sometimes without a human in the loop.

That difference changes what an enterprise data foundation must provide. Organizations that have built robust Gold-layer analytics environments are discovering a gap: their data is curated for human interpretation, not for machine reasoning. The business context that a skilled analyst carries in their head, such as what “customer” means across different systems, how revenue is calculated across departments, or which business rules govern a fulfillment decision, has never been formally encoded into the data architecture.

To address that gap, we use the term Platinum layer to describe a set of capabilities that add semantic intelligence, ontology-driven context, and governed AI enablement to the traditional medallion architecture. The result is a data foundation that supports enterprise AI at scale while preserving the governance and trust established through the medallion model.

In our work helping clients modernize data foundations, we often use Microsoft Fabric because it provides native, integrated support for the architectural capabilities discussed in this article within a single platform. According to Microsoft, Fabric serves more than 31,000 customers and is the fastest-growing data platform in the company’s history. At the time of writing, the capabilities described in this article are generally available or publicly announced by Microsoft rather than future-state concepts.

Why reporting-era architectures are no longer sufficient

Traditional analytics architectures were designed around a predictable consumption pattern: a human analyst queries structured data, a visualization tool renders the result, and a business leader interprets it. In this era, intelligence always lived at the edge: in the analyst’s domain knowledge, the report author’s framing, and the executive’s judgment.

AI systems change that model fundamentally. When an AI agent answers a procurement question, it doesn’t simply surface a pre-authored visualization. It traverses complex relationships across supplier, inventory, contract, and fulfillment data, applies business rules, and generates a response that may trigger downstream actions. Because of this shift, intelligence can no longer live at the edge; it must be embedded directly within the data architecture.

This shift creates three requirements that reporting-era architectures were not designed to address:

Cross-domain reasoning

AI systems must navigate relationships between disparate business entities (such as connecting a customer to an account, or an asset outage to a cold chain breach sensor) without human mediation. These relationships must be formally defined and structurally enforced, not assumed.

Semantic consistency

When two source systems define “active customer” or “net revenue” differently, a human analyst historically resolved the ambiguity through personal judgment. An AI system inherits that ambiguity and amplifies it at scale. Consistent semantic definitions must be baked into the architecture, rather than left as institutional knowledge shared between analysts.

Governed automation

AI-driven workflows, such as automated alerts, intelligent recommendations, and agentic orchestration, require data that carries deep provenance, ownership, and trust signals. A system that autonomously acts on data needs to know definitively whether it can trust the data it is acting on.

The medallion foundation

The three-layer medallion model remains the foundation for modern enterprise data platforms. Microsoft Fabric provides native, integrated support for each layer within a unified platform, reducing the multi-tool fragmentation that has historically complicated enterprise data management.

Figure 1: The basic medallion architecture

1. Bronze: Raw enterprise ingestion

The Bronze layer captures enterprise data in its original form. Its purpose is to preserve original source structure, auditability, lineage, and replayability. Nothing is discarded, and all data is traceable to its origin.

• OneLake Substrate: Serves as the unified storage substrate, a single logical data lake spanning the entire data estate.
• OneLake Mirroring: Allows supported source systems to replicate into OneLake without complex ETL orchestration, with additional sources accessible through Data Factory connectors and OneLake Shortcuts.
• Shortcut Transformations: Automatically converts ingested files to Delta format at the point of ingestion, making raw data immediately queryable while preserving source fidelity.

2. Silver: Harmonized, trusted data

The Silver layer transforms raw ingested data into a trusted, harmonized foundation that serves the entire enterprise. It is where data quality is established, structural inconsistencies are resolved, and cross-system records are aligned. Through cleansing, standardization, integration, enrichment, deduplication, and the normalization of semi-structured data, this layer produces reusable, governed enterprise data sets that can be shared and trusted across business domains.

• Data Factory: Orchestrates the end-to-end transformation workflow, ensuring cleansing, standardization, enrichment, and deduplication execute reliably and in the right sequence.
• Spark Notebooks: Handle complex enrichment and deduplication logic at enterprise scale, giving data engineers full programmatic control where low-code tools reach their limits.
• Dataflows Gen2: Provides a low-code visual interface for transforming and standardizing data across sources, making data preparation accessible beyond engineering teams.

3. Gold: Curated business analytics

The Gold layer produces the business-ready data products that drive traditional enterprise analytics: certified KPIs, dimensional models, curated reporting datasets, and governance-enforced data sharing across business domains. This is where data becomes useful for human decision-making.

Core activities include dimensional data modeling (slowly changing dimension (SCD) patterns), star schema construction (fact and dimension tables), business-driven transformations (KPI logic and aggregation rules), and performance optimization via materialized views to support fast, self-service analytical consumption. It applies certification, lineage, and ownership metadata so consumers know what they can trust while enforcing secure data sharing access policies across business domains.

• Fabric Data Warehouse: Serves as the unified storage substrate, a single logical data lake spanning the entire data estate.
• OneLake Data Sharing: Allows supported source systems to replicate into OneLake without complex ETL orchestration, with additional sources accessible through Data Factory connectors and OneLake Shortcuts.
• Dataflows Gen2: Applies business-driven transformations and KPI logic natively within Fabric, enabling analysts to build and maintain curated datasets without requiring engineering resources.

The Platinum layer: Enterprise intelligence for the AI era

The Platinum layer is a structured set of capabilities that create a governed semantic environment. Here, business meaning, entity relationships, contextual metadata, AI-ready interfaces, and intelligent agent capabilities are formally defined, governed, and made reusable across every consumer of enterprise data.

In Microsoft Fabric, many of the capabilities that support the Platinum layer are delivered through Fabric IQ, which provides a common framework for defining, governing, and consuming business context across enterprise data.

Figure 2: The Platinum layer creates a governed semantic environment.

The Ontology: Encoding the language of the business

At the center of the Platinum layer is the Ontology, Fabric IQ’s foundational capability for defining the shared language of the enterprise. It answers the questions that data models alone cannot: not just which tables exist, but what a “customer” actually means across finance, operations, and fulfillment, or how revenue is calculated consistently across business units. Once defined, the Ontology becomes the authoritative reference that AI systems, analysts, and automated workflows all draw from, eliminating the ambiguity that has historically lived between teams.

Figure 3: Fabric IQ and its components

In Fabric IQ, the Ontology can be built from scratch or derived from existing semantic models and data assets already in the platform. By grounding the Ontology in existing enterprise assets, organizations can create a governed business vocabulary that helps ensure AI systems reason from consistent definitions rather than conflicting interpretations.

Graph: Understanding how the business connects

The Graph capability in Fabric IQ maps and traverses the relationships between business entities defined in the Ontology. Where traditional data models focus on individual records, the Graph answers questions about how things relate: which suppliers are connected to a delayed shipment, which assets are linked to a compliance risk, which customers sit at the intersection of two revenue streams. This connected intelligence is what allows AI systems to reason across the enterprise rather than within a single domain.

Plan: Connecting data to business decisions

The Plan capability in Fabric IQ brings collaborative planning, budgeting, and forecasting directly into the governed data environment. Business users work from the same semantic definitions and certified metrics used across the platform, reducing the disconnect between planning activities and enterprise data. Every planning action carries a full audit trail, ensuring accountability without sacrificing agility.

Fabric Data Agents and Operations Agents: AI that acts on trusted data

Fabric Data Agents are the AI-facing interface of the Platinum layer. Rather than relying solely on unstructured prompts, they reason from the Ontology, certified metrics, and governed data products built across Bronze through Gold layers. An executive asking about supply chain exposure or customer retention risk receives an answer grounded in the same definitions and business rules that govern every other enterprise decision.

Operations Agents extend this capability into proactive intelligence. They continuously monitor enterprise data and automatically trigger defined actions when business conditions are met, whether that is flagging an anomaly, initiating a remediation workflow, or escalating to the right team. Every action is traceable back to the business rule that authorized it, making AI-driven automation auditable by design.

Power BI semantic models: One definition, every consumer

Power BI semantic models provide a governed metric layer within the Platinum layer, defining certified KPIs, business hierarchies, and calculation logic that every consumer, whether it’s a Power BI report or a Fabric Data Agent, draws from identically. When an AI agent answers a revenue question, it uses the exact same certified definition that the CFO’s dashboard uses. Consistency is enforced by the architecture, not by convention or individual judgment.

Governance: The connective tissue across all layers

Governance is not a separate layer in this architecture. It spans Bronze through Platinum, providing the trust, ownership, lineage, and accountability that make every other layer usable. Without governance, the Platinum layer cannot fulfill its purpose: a semantic model without certified ownership drifts, an ontology without stewardship becomes a source of inconsistency, and an untracked AI agent produces untrustworthy outputs.

Figure 4: Governance spans all layers from Bronze through Platinum.

• OneLake Security: Provides unified role-based access control (RBAC) that spans all consumption paths (analytical queries, API access, agentic workloads) without requiring separate security configurations per tool. Row and column-level security are enforced natively across all workloads, including real-time operational data via Eventhouse.
• OneLake Catalog (Govern Tab): Provides domain and workspace-level governance insights, including ownership, certification status, lineage, sensitivity classifications, and auto-generated semantic descriptions-in a single view.
• Agentic Discovery: The native Fabric MCP server enables AI systems to locate, evaluate, and consume certified data products programmatically through the same governance framework.

For AI workloads, every insight must be traceable back to a specific certified data product, ontological definition, and version of the business rules. This traceability helps organizations understand where AI-generated insights come from, which business definitions informed them, and whether they can be trusted in operational decision-making.

A four-layer reference architecture

Where enterprises should start

The scale of this four-layer reference architecture can make it seem like a massive, multi-year transformation program. However, the most successful modernizations begin with a single, high-value decision domain to establish the patterns and organizational capability needed to scale.

The 5-step approach

1. Establish the medallion foundation in Fabric first

Build Bronze through Gold for one high-priority use case, ensuring data quality, lineage, certification, and ownership are established before introducing the Platinum layer.

2. Choose a use case with strong AI opportunity

Focus on a use case where AI reasoning across multiple entities produces immediate business value, such as customer operations, supply chain visibility, asset reliability, regulatory reporting, or financial planning.

3. Define the ontology for that use case

Collaborate with data architects and domain subject matter experts to map out the core entities, relationships, and business rules using the language the business actually uses.

4. Implement Fabric IQ for that use case

Connect your existing semantic models and data sources to the Fabric IQ workload, certify core metrics, and assign stewardship so AI systems have a governed, trusted foundation to reason from.

5. Deploy one Data Agent grounded in the Platinum layer

Build a scoped, domain-specific virtual analyst that reasons from the ontology and semantic model to demonstrate what enterprise AI means in practice.

Three critical failure modes to avoid

Enterprise-wide ontology design before any domain is production-ready

Semantic transformation is an iterative capability. Attempting to design an enterprise ontology all at once creates a massive design program with no clear end point. Start with one use case, prove the pattern, and scale.

Building AI features on Silver-layer data

The Platinum layer must sit directly on top of Gold, not Silver. AI agents built on unharmonized, non-curated data inherit every structural inconsistency and will surface those errors in their responses.

Treating governance as a downstream step

Governance retrofitted onto a completed architecture does not hold. Lineage, ownership, and certification must be integrated into the architectural design from the Bronze layer forward.

The foundation for enterprise AI begins here

The organizations that will lead the AI era are not necessarily those with the most data, but those with the most semantically coherent, governed data foundations. Microsoft Fabric provides an integrated platform to build this end-to-end.

While the Gold layer successfully transformed human enterprise analytics, the Platinum layer is what will transform the enterprise itself. By embedding business context, governance, and semantic intelligence directly into the architecture via Fabric IQ, organizations are building the infrastructure that makes AI trustworthy, auditable, and genuinely useful for the long haul.

Learn more about enterprise AI data foundations