The Architecture AI Demands: A Perspective Informed by Industry Research on ABI Architectures

Enterprise architecture is being rewritten in front of us.
Not by vendors. Not by standards bodies.
But by the practical realities of AI adoption inside large organizations.

For years, data teams focused on building warehouses, pipelines, dashboards and governance frameworks. These were necessary—but they were never designed for a world where every function expects AI-powered reasoning, conversational interfaces and real-time decisions grounded in enterprise truth.

Now, AI is forcing a structural rethink.

Recent research from Gartner underscores “Organizations continue to adopt tools that enable novice and professional analytics developers to deliver insights and facilitate decision making. To succeed, analytics specialists must implement the architectural principles and capabilities that enable analytics and business intelligence (ABI) use cases.”

From Kyvos’ perspective, the growing emphasis on semantic modeling reflects how AI has amplified the cost of inconsistency and increased the value of shared, governed meaning across analytics and AI use cases.

In this article, I want to lay out—purely from a business leader’s lens—what the new enterprise architecture looks like, why a semantic layer becomes the “control plane” for AI+BI and what this means for enterprises navigating the next decade.

The AI Era Exposes the Limits of Traditional Architecture

For the last 20 years, enterprise data architecture followed a predictable pattern:

• Collect data
• Clean it
• Store it
• Visualize it
• Govern it

This worked when analytics meant dashboards and KPIs.

But today, leaders expect more:

• AI that understands business context
• AI that doesn’t hallucinate or contradict dashboards
• Insights that work across tools, functions and personas

Traditional architectures were not built for this.

They optimized for storage and compute, not for meaning.

AI, especially Gen AI, flips this equation.

Enterprises now need a layer that makes data interpretable, consistent and reusable—independent of tools or interfaces. According to us, this is why Gartner’s architecture places semantic modeling as a central capability—on equal footing with data preparation, visualization, ML and BI.

A New Enterprise Architecture Is Emerging

If you look at the reference model, the evolution becomes clear:
The modern architecture is no longer linear; it is compositional, modular and semantically governed.

The new stack looks like this:

Layer 1 — Data Foundation (Warehouses, Lakes, Fabrics)

Still essential but now treated as “table stakes” infrastructure: scalable, flexible, cloud-native.

Layer 2 — Data Preparation + Transformation

Automated, ML-assisted cleaning, profiling

Layer 3 — Semantic Modeling (The New Control Plane)

This is the most important shift.
Semantic modeling now defines:

• Business entities
• Metrics and logic
• Hierarchies
• Relationships
• Data access policies
• Cross-domain meaning

This is where consistency begins.

This is also the layer as essential for AI grounding, BI and unified interpretation across tools.

Layer 4 — Augmented Analytics + Machine Learning

Predictive, prescriptive and generative capabilities that rely on semantic consistency to avoid misinterpretation or drift.

Layer 5 — Visualization, Delivery and AI Interfaces

Dashboards, applications, chatbots, copilots, embedded analytics—anything that touches the end user.

Layer 6 — Analytics CI/CD & Governance

A structured lifecycle for versioning, publishing and monitoring analytics and AI assets.

This new framework reflects a philosophical shift:

Meaning and governance are becoming upstream artifacts, not downstream fixes.

Why Semantics Is Suddenly a Boardroom Issue

Historically, semantics were seen as a BI construct—something to help dashboard teams maintain consistency.

But AI changes the stakes.

AI systems interpret raw data, documents, metrics and unstructured content. Without a shared semantic layer, each system forms its own interpretation—a breeding ground for drift, contradiction and hallucination.

Gartner report Reference Architecture Brief: Analytics and Business Intelligence explicitly states:

The reference architecture to enable ABI focuses on an infrastructure with seven core components that support scalable, flexible and governed citizen development. Those components include:

• Data preparation — Tools and services that give users the ability to connect with and prepare refined data for analytical modeling.
• Semantic modeling — Core capabilities for modeling data and creating metrics for analytics use cases.
• Augmented analysis — Capabilities that enhance the analytics development process and provide automated insights through machine learning (ML) and generative AI (GenAI) capabilities.
• Visualization and delivery — Tools that allow analytics users to analyze, visualize, and ultimately share their insights by publishing and sharing reports and dashboards. Advanced deployment capabilities may include embedding analytics in application workflows and building conversational interfaces.
• Analytics continuous integration/continuous delivery (CI/CD) experience — Tools for managing the deployment processes and source code control for analytics assets.
• Analytics operations — The control plane for analytics solutions that provides the capabilities to configure, manage and monitor the architecture.
• Analytics governance —Tools for inventorying, securing, promoting and monitoring assets in the architecture.

From Kyvos’ perspective, these capabilities collectively highlight why semantic modeling has become essential for consistency, reuse, and reliable AI-driven analytics across tools and personas.

In other words: Semantics is not a BI feature anymore. It is a “trusted, governed AI”-enablement layer.

And for the first time, it impacts every persona:

• CIO/CTO → reduced architectural fragmentation
• CDAO → consistent governance and trust
• Analytics leaders → no metric disputes
• AI leaders → grounded, explainable AI
• Business teams → aligned metrics, faster adoption

The Semantic Layer as the “Enterprise Memory” for AI

If AI is the reasoning engine, the semantic layer is the memory it reasons on. We are moving toward architectures where:

• AI copilots query semantic models, not raw schemas
• BI and AI share the same definition for revenue, churn, margin
• Agents navigate enterprise knowledge using standardized entities
• Data products are built on reusable semantic definitions
• Interpretation becomes consistent across tools, functions, and geographies

The architectural implication is clear: semantics are no longer confined to reporting layers but must operate across analytics, AI, and emerging agentic workflows.

What This Means for Enterprise Leaders

From my perspective—after writing and thinking deeply about real-world AI adoption—the implications are clear:

AI readiness is no longer about having data. It’s about having meaningful data.

Tool consolidation is less important than semantic consolidation.

AI value creation depends less on models and more on shared definitions.

Governance must shift from reactive correction to proactive semantic standardization.

Architectures must support conversational, agentic and analytics use cases simultaneously.

In short:

Enterprises don’t just need modern data platforms.

They need modern semantic platforms.

A Practical Framework: How Leaders Should Respond

Here’s a business-friendly blueprint for adopting this new architecture:

1. Treat the semantic layer as a first-class architectural asset

• Not an optional BI feature.
• Not a reporting convenience.
• A strategic enterprise layer.

2. Implement semantics in phases, but unify them enterprise-wide

• Begin with high-value business areas (e.g., Finance, Customer, Supply Chain).
• Establish foundational semantic definitions—entities, metrics, hierarchies.
• Expand into adjacent domains while preserving shared meaning.
• Converge these areas into one enterprise semantic model, not siloed domain models.

3. Unify metrics and definitions across AI and BI from Day 1

• Identify enterprise-critical KPIs (revenue, churn, margin, etc.).
• Define them once in the semantic layer—not separately in BI tools or AI prompts.
• Ensure both dashboards and AI assistants use the same governed definitions.
• Continuously evolve metrics centrally rather than in tool-specific layers.

4. Expand the semantic model as new business questions emerge

• Treat new AI/analytics use cases as opportunities to enrich the semantic layer.
• Add new entities, hierarchies and rules to the enterprise model—not one-off solutions.
• Map new data sources to existing semantics to avoid metric drift.
• Ensure all downstream tools inherit these updates automatically.

5. Make interoperability a design principle, not an afterthought

• Ensure the semantic model is accessible through APIs, SQL, MCP, etc.
• Map semantics once, reuse them across Power BI, Tableau, Looker, AI chatbots and agents.
• Let business units choose tools while remaining aligned on shared definitions.
• Avoid embedding semantics inside tools where they become brittle and fragmented.

6. Treat governance as an accelerator, not a gatekeeper

• Move governance “upstream” by governing definitions, not dashboards.
• Centrally manage logic, hierarchies and access rules through the semantic layer.
• Allow teams to build faster because definitions are already standardized.
• Use governance to prevent BI and AI drift, not to block development.

7. Integrate semantics directly into AI workflows (not as an overlay)

• Plug LLMs, AI chatbots and agents into the semantic layer, not raw data.
• Use semantic entities and metrics as AI’s “enterprise vocabulary.”
• Ensure retrieval, prompt orchestration and contextual augmentation pull from governed semantics.
• Validate AI answers against semantic definitions automatically.

8. Build analytics and AI products on top of shared semantic assets

• Treat semantic entities and metrics as composable components.
• Allow teams to assemble dashboards, data apps and AI workflows using shared semantics.
• Reduce duplication by centralizing logic in one place instead of recoding it in every project.
• Use semantic consistency to scale insight delivery across business units.

How Kyvos Embodies This New Semantic-First Architecture

As enterprises shift toward this new architecture—where semantics anchor trust, accelerate analytics and ground AI—Kyvos naturally fits into the center of this transformation.

Kyvos was built long before the Gen AI wave with a foundational belief: If intelligence drives decisions, then meaning must drive intelligence.

Today, that belief aligns precisely with what Gartner’s evolving architectures emphasize: enterprise semantics, cross-tool consistency, governed metrics and scalable performance.

Kyvos delivers:

1. A unified data foundation for AI + BI

Centralized definitions across Power BI, Tableau, Excel, AI copilots and go ro metric drift. Zero contradictions. One enterprise truth.

2. Enterprise-scale performance

Sub-second queries on massive datasets, enabling interactive analytics and feeding AI models with governed, high-speed context.

3. A bridge between legacy analytics and modern AI

Kyvos modernizes legacy platforms, unifies fragmented definitions and preserves historical investments—while elevating them into a semantic-first future.

4. AI grounded in enterprise truth

Kyvos reduces hallucinations and contradictions by grounding AI in governed semantic meaning.

5. Composable semantic assets

Metrics, hierarchies and entities become reusable components for faster insight delivery.

6. Tool and cloud interoperability

Works across BI platforms, data platforms, clouds and AI ecosystems—without lock-in.

7. Governance designed for scale

Upstream governance embedded directly into the semantic layer: access policies, lineage, versioning and secure metric management.

Kyvos is not just a BI accelerator.

It is the semantic layer that makes enterprise AI real, reliable and scalable.

The Bottom Line

We are entering an era where AI becomes the default interface for enterprise knowledge.

For AI to succeed, your data must be:

• Consistent
• Interpretable
• Explainable
• Reusable
• Aligned across tools and teams

That is the promise—and necessity—of the semantic layer.

In our opinion, Gartner reference architecture reflects how modern analytics and BI capabilities are being structured across the industry.

This broader industry evolution makes semantic consistency increasingly difficult to treat as optional. And for enterprises ready to modernize, this is the clearest path to scalable, trusted, real-world AI.

Gartner, Reference Architecture Brief: Analytics and Business Intelligence, Christopher Long, 20 October 2025

GARTNER is a trademark of Gartner, Inc. and/or its affiliates.