# How do you model enterprise architecture from structured and unstructured sources with AI?

> Canonical: https://decisionpen.com/blog/how-to-model-enterprise-architecture-from-structured-and-unstructured-sources-with-ai  
> Published: 2026-10-27 · DecisionPen · Automated Landscape Modelling series (D7b)  
> Related docs: /docs/incremental-modelling, /docs/architecture-decision-workspace, /docs/beyond-traditional-ea, /docs/enterprise-ask

## Summary

Use structured sources (CMDB or inventory spreadsheet) to establish identity — what applications and technology components exist — and unstructured sources (HLDs, wikis, transcripts, natural language) for meaning: purpose, integrations, constraints, rationale. AI proposes metamodel-aware entities and relationships from both; evidence is bound; architects curate incrementally. Neither source suffices alone.

## Structured vs unstructured

- **Structured:** CMDB export, application inventory spreadsheet, technology catalogue — strong on what exists, IDs, owners, lifecycle fields.  
- **Unstructured:** HLDs, solution documents, wiki pages, transcripts, decision emails, natural-language descriptions — strong on how and why.

## Why each fails alone

- Structured only: tidy list, no How/Why; impact analysis is guesswork.  
- Unstructured only: inconsistent vocabulary (“the gateway” vs “PGW”), duplicate proposals, curation becomes deduplication.  
Together: structured gives AI names to resolve against; unstructured gives entities something to say.

## Workflow

1. Scope one domain.  
2. Seed identity by importing the structured extract (applications, technology components).  
3. Ingest unstructured evidence for the same domain.  
4. AI proposes against the metamodel — matching to existing entities, flagging new ones.  
5. Bind evidence (source passage/row).  
6. Architects accept, edit, reject, or defer.  
7. Repeat with the next document or sprint.

## Incremental, not one baseline

A baseline is a snapshot with a half-life. Incremental modelling treats each meeting, document, and review as a small update to a living map. Progress = trusted updates, not % of estate covered. A curated slice of one domain can already ground Ask and artefacts.

## Product boundary (be explicit)

“Structured” today means **spreadsheet import**: export the CMDB/inventory you maintain and import applications, technology components and similar items as a skeleton. It is **not** a live, bi-directional ServiceNow (or other CMDB) sync. For continuous operational discovery or two-way CMDB sync, use a different tool; DecisionPen is the curated architecture layer on top.

## Common failure modes

Duplicate entities (seed identity first, merge early); over-eager relationships (mention ≠ integration; check evidence); stale sources (record which claim wins); unbounded ingestion (scope by domain, add in waves); treating AI output as fact (curate).

## LeanIX alignment

Proposals can be aligned to LeanIX entity types where that is the organisation’s modelling language.

## How DecisionPen helps

Spreadsheet import for the structured skeleton; document and dialogue ingestion; metamodel-aware proposals with bound evidence; human curation; Enterprise Ask and artefacts from the curated map. Trial: https://decisionpen.com/trial

## When is DecisionPen the wrong tool?

Live bi-directional ServiceNow/CMDB sync, network/SaaS scanning, Visio/Lucid drawing, BPMN, or Copilot-only text drafting as the primary need.

## FAQ

**How do you model EA from structured and unstructured sources with AI?**  
Seed identity from a CMDB/inventory spreadsheet, ingest documents and dialogue for meaning, let AI propose metamodel-aware entities and relationships with evidence, and curate incrementally.

**Why not use only a CMDB?**  
It tells you what exists, not capability framing, integrations, or rationale.

**Why not use only documents?**  
Inconsistent naming produces duplicates and no stable identity; a structured skeleton anchors proposals.

**Does DecisionPen sync live with ServiceNow?**  
No. Today it supports spreadsheet import of CMDB/inventory extracts, not live bi-directional sync.

**How does DecisionPen help?**  
It combines spreadsheet import, document/NL ingestion, metamodel-aware proposals, evidence binding and curation in one Architecture Decision Workspace.

**When is DecisionPen the wrong tool?**  
For live CMDB sync, scanning, freeform drawing, BPMN, or Copilot-only drafting as the sole need.

## Series index (D7)

- https://decisionpen.com/blog/what-does-ai-do-inside-an-architecture-decision-workspace  
- https://decisionpen.com/blog/how-to-model-enterprise-architecture-from-structured-and-unstructured-sources-with-ai  
- https://decisionpen.com/blog/how-should-humans-review-and-curate-ai-proposed-architecture-models  

## Related posts

- https://decisionpen.com/blog/automated-it-landscape-modelling-from-unstructured-data  
- https://decisionpen.com/blog/how-ai-accelerates-automated-enterprise-discovery  
- https://decisionpen.com/blog/how-to-enrich-ea-factsheets-with-unstructured-documents  
- https://decisionpen.com/blog/automatically-generate-leanix-compliant-enterprise-baseline  
- https://decisionpen.com/blog/top-down-vs-bottom-up-enterprise-architecture-modelling  
- https://decisionpen.com/blog/why-ea-factsheets-miss-how-and-why