An architecture position pack

Enterprise Data Architecture

For CIOs, CDOs, enterprise architects, and platform leaders designing the boundary between analytical and operational systems.

Every enterprise builds a data platform. Most of them also accidentally try to make it run their business operations. This is the guide to not doing that: hard boundaries between enterprise data platforms and operational platforms, decision frameworks, and reference blueprints with real cloud provider mappings.

One flow, three responsibilities. The enterprise data platform is the system of insight; the expensive failures start when it is asked to be the other two.

This pack covers one of enterprise data architecture's most consequential decisions: where systems of record, insight, and action should remain separate, and where they may safely share infrastructure. The position it argues is governed coexistence: every workload gets an explicit architectural contract, and logical responsibility boundaries survive even where the technology converges. It advocates responsibility boundaries, not technology fragmentation, and it is deliberately not a complete treatment of every data-management discipline.

40 pages · 7-layer architecture · 35+ capabilities · 4 compliance regimes · agent access patterns · 3 case studies · 5 review checklists

Edition 2 · Updated July 2026

The Guide

Position

What the enterprise data platform is, what it is not, and why the boundary matters.

6 pages

Blueprints

Reference architectures: seven layers, capabilities, control plane, AI supply chains, agent access.

7 pages

Scenarios

Worked example architectures: an EU insurer on Azure, a tier-1 bank on GCP, a US healthcare provider on AWS.

3 pages

Decisions

Workload routing, capability maps, data mesh, maturity, and vendor evaluation frameworks.

6 pages

Patterns

Data contracts, cost architecture, open table formats, and the catalog layer.

3 pages

Compliance

BCBS 239, DORA, HIPAA, Solvency II, IFRS 17, and the EU AI Act mapped to platform design.

5 pages

Operations

Roles, processes, KPIs, support tiers, SLOs, and incident discipline for running the platform.

2 pages

Transformation

A four-stage roadmap from current-state confusion to governed coexistence.

1 page

Evidence

Worked examples, evidence tables, target metrics, and review checklists, each labeled by provenance.

5 pages

Glossary

20 commonly confused terms, each mapped to the target-state architecture.

1 page

Start Here

I need to explain platform boundaries to leadership

What EDP Is The problems an enterprise data platform solves and the problems it must not solve
EDP vs Operational Platform Side-by-side comparison across 12 dimensions, plus the 2026 convergence argument
Anti-Patterns What breaks when EDP becomes everything
EDP Is Not an ODS The most common misuse, the five differences that bite, and how to spot it
The New Data Architecture The four 2026 shifts worth designing for, and what did not change
Convergence and Exceptions When analytical platforms may legitimately serve operational reads, and the contract test that decides

I need a reference blueprint

Target-State Architecture Seven-layer reference architecture with GCP, Azure, and AWS service mappings
Capability Architecture 35+ capabilities the EDP must provide, with components mapping
Control Plane Metadata, lineage, policy, contracts, audit, and observability infrastructure
AI/ML Platform Relationship The classic ML and LLM data supply chains, from feature stores to RAG corpora
Agents as Consumers Governed agent access through MCP, semantic layers, and scoped identity
Coexistence Patterns How EDP and operational platforms connect through five integration patterns
The Serving Layer The four serving shapes, feeding patterns, and where routed workloads actually go

I need a worked example

Insurer on Azure Solvency II, IFRS 17, and AI Act pressure; an accidental ODS unwound; a governed claims copilot
Bank on GCP BCBS 239 and DORA; unwinding a one-platform mandate; the risk-report lineage walk
Healthcare on AWS HIPAA zones, de-identification placement, tokenized linking, research data products

I need to make an architecture decision

Decision Tree "Where does this workload belong?" flowchart with quick reference table
Business Capability Map 15 business capabilities mapped to platform owners
Data Mesh When data mesh works, when it does not, and the pragmatic middle ground
Maturity Model Five levels across six dimensions
Vendor Evaluation 10-dimension evaluation framework with platform archetypes
Capability Assessment 15 capabilities x 4 maturity levels assessment

I need integration patterns

Data Contracts Schema, SLAs, ownership, and evolution rules between producers and consumers
Cost Architecture FinOps patterns for data platforms
Open Formats and Catalogs Iceberg as the interchange layer, catalog selection, zero-ETL, and the streaming lakehouse

I need regulatory compliance guidance

Compliance Overview Cross-industry regulatory matrix mapped to platform design
Banking BCBS 239 and DORA mapped to architecture decisions
Healthcare HIPAA mapped to data platform design
Insurance Solvency II and IFRS 17 data requirements
EU AI Act Article 10 data governance and Article 12 logging mapped to platform capabilities

I need an operating model

Operating Model Roles, processes, measures, and support tiers for running the EDP
Reliability Model Platform SLOs, incident classification, recovery patterns

I need a transformation plan

Transformation Roadmap Four stages from current-state confusion to governed coexistence

I need evidence and applications

Case Studies Three illustrative composite transformations
Evidence Tables Claims, counter-arguments, failure modes, and what good looks like, labeled by provenance
Decision Records Worked ADR examples applying these frameworks
Metrics and Outcomes Target ranges across 13 dimensions, with what to measure
Review Checklists Five operational checklists for architecture review boards

About the Author

Sunil Prakash is an enterprise AI and data platform architect working in global banking. His work spans governed data platforms, cloud transformation, platform governance, and AI and agent architecture; the frameworks in this pack come from applying and defending these boundaries inside large regulated enterprises. He writes on agent infrastructure at the Agent Engineering Lab and publishes implementation companions to this guide on GitHub. For architecture reviews, advisory conversations, or speaking: sunil@sunilprakash.com.

Cite as: Prakash, Sunil. Enterprise Data Architecture: A Position Pack for Governed Coexistence. 2026. sunilprakash.com/enterprise-data-architecture.

Related Implementation

This guide is the strategy layer. For working implementations: reference-data-platform-gcp (production-grade EDP on GCP: Data Vault 2.0, dbt, BigQuery, Terraform) and dbt-data-vault-starter (an opinionated dbt template for Data Vault 2.0). Unclear on a term? Start with the glossary.