AI and the Future of Data Modelling

AI won’t replace data modelling - but it will fundamentally change how we do it.

Used well, AI can accelerate application data modelling, improve consistency and reduce friction across delivery teams. Used badly, it simply produces convincing-looking diagrams that don’t align to enterprise meaning or governance.

The difference isn’t the AI - it is the foundations we give it.

Why AI struggles with data modelling today

Have you ever asked an AI tool to create a data model?

If you have, you’ve probably seen the same thing most people do: something that looks plausible at first glance, but isn’t deployable - and almost never reflects your organisation’s shared data language.

That isn’t a failure of AI. It’s a lack of context.

AI is only as good as the information it has to work with. While some data models exist in the public domain, very few are genuinely high quality. Most robust industry reference models are vendor IP - and priced accordingly.

So if organisations want to use AI effectively for data modelling, they can’t rely on generic models scraped from the internet. AI needs to be grounded in their enterprise language, rules, and structure.

That’s why the real opportunity lies in AI models trained specifically for data modelling - models that understand enterprise definitions, constraints, model types and modelling patterns, not just syntax.

But there’s an important prerequisite.

AI doesn’t replace modelling discipline - it amplifies it

Before AI can accelerate data modelling, we need to understand the modelling process ourselves.

AI doesn’t remove the need for good modelling discipline. It amplifies it.

If your process is unclear, AI will simply help you reach the wrong answer faster.

To make this concrete, I’ll use the data model hierarchy from my previous article, “Evolution of the Data Model Hierarchy.” If you haven’t read it, here’s a quick recap.

Where AI fits in the modelling process

The diagram below shows a simple “happy path” for delivering an application data model.

If AI is going to help, it needs to fit into this flow - accelerating good practice rather than bypassing it.

The goal isn’t to replace the modelling process. It’s to embed AI into it so good practice becomes faster and more consistent by default.

One thing is worth being explicit about: AI does not replace data modelling tools.

We will always need modelling tools to:

• hold the master copy of the model

• support collaboration, versioning and branching

• enforce governance and traceability

AI sits alongside these tools, accelerating decisions without bypassing control. That means data models must flow cleanly to and from AI in a governed way.

What AI needs to be effective

For AI to create useful application data models, it must be trained on an organisation’s own data foundations, including:

• Business Glossary, Ontology & Rules

• Enterprise Data Models (CDM & ELDM)

• Application data model types

• Application modelling patterns (history, lifecycle, auditing, versioning, temporality, etc.)

Without this grounding, AI is likely to hallucinate - confidently.

How AI can accelerate forward engineering

With strong foundations in place, AI can support - and accelerate - each step of the forward-engineering process without breaking governance.

1. Create/Update Business Concept Model  Diagram
   Using semantic matching, AI can analyse application requirements and map them to existing items in the Business Glossary, Ontology, and Rules.
   From this, it produces a Business Concept Model diagram representing the subset of concepts relevant to the application. The diagram, linked to the underlying ontology elements, is saved to the modelling tool where the modeller can refine it.

2. Create/Update EDM Diagrams (CDM & ELDM)
   AI generates a Conceptual Data Model using traceability from the Business Concept Model. It then produces the Enterprise Logical Data Model, deriving entities from the CDM and selecting attributes based on linked business concepts.
   Where appropriate, AI can suggest additional attributes, leaving the final decision with the modeller. Both diagrams are saved to the modelling tool with full traceability intact.

3. Determine Application Data Model type
   Based on application requirements — such as purpose (operational vs analytics), query patterns, data scope, history, and audit needs — AI can recommend the most appropriate application data model type, as described in Types of Application Data Model. 

4. Select Application Data Modelling patterns
   Traditionally, these patterns would be determined by the modeller as part of the modelling process. In an AI-assisted approach, patterns such as lifecycle management, versioning, temporality and auditing are provided as context. AI selects the relevant patterns based on requirements and presents options where multiple approaches are valid.

5. Create/Update ALDM
   With all required context available, AI generates the ALDM using:

• the relevant subset of the ELDM

• the chosen application data model type

• the selected modelling patterns

The resulting model is saved to the modelling tool with full traceability to enterprise data model elements. The modeller then refines and validates the design.

Note: AI does not need to create the Physical Data Model. Existing modelling tools already handle physical implementation effectively. 

  !  Avoiding Delivery Bottlenecks

Where new business concepts or data elements are required, AI can create placeholders in a feature branch. This avoids enterprise modelling teams becoming delivery bottlenecks while preserving lineage. Those changes can then be added automatically to a backlog for review, formalisation and controlled merge - without losing audit history.

Additional ways AI can help

Beyond model creation, AI can also act as a second pair of eyes:

• reviewing models against naming, standards and traceability rules

• generating quality assessments and governance reports

• highlighting gaps before implementation, not after

Conclusion

AI won’t magically fix poor data modelling.

But combined with strong enterprise foundations, it can dramatically improve how quickly - and consistently - good application data models are delivered.

The future of data modelling isn’t about replacing architects or tools. It’s about using AI to carry context, enforce discipline, and remove unnecessary friction - so humans can focus on judgement, trade-offs and design intent.

In short: AI can help us model faster - but only if we already know how to model well.