How To Start A Data Model

For many data professionals, the hardest part of data modeling is not the technical work.

It’s getting started.

You sit down with a blank sheet of paper and a request to “produce a data model.” At that moment, the challenge is figuring out where to begin.

In a recent Business Thinking Podcast discussion, we explored a practical approach to starting a data model from scratch—especially when working with Data Vault.

The goal is not to produce a perfect model immediately. The goal is simply to get the first ideas on paper and begin refining them.

Start by Asking Questions

One of the simplest ways to begin modeling is also the most obvious: start asking questions.

Instead of trying to design the entire structure upfront, begin by talking to people who understand the business area you’re working in.

Ask questions such as:

• What happens in this part of the business?

• What steps are involved in the process?

• What things are involved in those steps?

Even rough conversations will start to reveal key ideas. As those ideas appear, you can begin connecting them.

In practice, this often looks like a simple mind map of concepts and relationships. Even if it doesn’t look like a formal model yet, you have already begun building one.

Focus on Processes First

If starting directly with a data model feels difficult, start somewhere more familiar: the business process.

Business users usually think in terms of processes rather than data structures. By describing how work actually happens, the important data concepts naturally appear.

A process usually involves:

• Actions

• People

• Objects or things being acted on

Those “things” often become the key concepts in the model.

For example:

• A business process might involve taking an order from a customer.

• Immediately, two concepts emerge: customer and order.

• If the order includes items being sold, then product may appear as another concept.

These concepts can then become the starting point for the model.

From Concepts to Data Vault Structures

When using Data Vault modeling, this process becomes especially clear.

In Data Vault terms:

• Concepts become hubs

• Relationships between concepts become links

For example:

• Customer → Hub

• Order → Hub

• Product → Hub

If customers place orders, that relationship becomes a candidate for a link.

At this stage, the goal is not to define everything perfectly. The goal is simply to capture the main ideas so they can be refined later.

Use Simple Tools First

Many modeling tools exist, but they are not always the best place to start.

Jumping straight into a modeling tool can sometimes lock your thinking too early. Once objects are placed inside software, people tend to hesitate about changing them.

A more flexible approach is to start with something simple, such as:

• A whiteboard

• Paper and pen

• Sticky notes

Concepts can be written on notes and moved around easily as the model evolves. Relationships can be sketched with simple lines.

Some teams also use digital whiteboard tools such as Miro when working remotely.

The important point is that early modeling should stay flexible and exploratory.

Generative AI Can Help You Get Started

Another useful way to break through the “blank page” problem is to use generative AI tools.

If you are unsure what concepts might exist in a particular business area, you can ask an AI system questions such as:

• What processes exist in this industry?

• What key concepts are involved in this type of business?

Providing context improves the results. For example:

• Specify the industry

• Mention the business function

• Include the country if regulations differ by region

The output can then act as a starting point for discussion and refinement, rather than a final answer.

Expect Multiple Versions of the Model

Early models are rarely correct.

In practice, a concept model may go through many revisions before it stabilizes.

Each revision clarifies the meaning of terms and relationships.

As the model evolves:

• Definitions become more precise

• Concepts may be renamed

• Relationships become clearer

This process is normal. Modeling improves through iteration and discussion.

Making mistakes along the way is not a problem—it is part of the learning process that leads to a stronger model.

Keep the Scope Focused

Another common mistake when starting a data model is trying to model too much at once.

Instead, focus only on the problem you are currently solving.

If you are modeling a customer order process, you do not need to fully model the finance department at the same time.

Related concepts may appear, but they can remain outside the scope of the current model until they become relevant.

A useful guideline is simplicity: If a model grows larger than a single page, it may be time to split it into multiple views.

Turning Early Ideas Into a Structured Model

Once the early concepts and relationships are identified, the structure becomes clearer.

At that point:

• Concepts can become hubs

• Relationships can become links

• Additional detail can be added gradually

Over time, the model becomes more stable as the system grows and new areas of the business are integrated.

The key lesson is that data modeling rarely starts with a perfect design. It starts with a rough sketch that improves through collaboration and iteration.

The First Step Is Simply Starting

Many professionals hesitate because they feel they must produce a perfect model from the beginning.

In reality, the most important step is simply to start drawing something.

Talk to people. Map the business process. Write concepts on a whiteboard.

Once those first ideas appear, the model begins to take shape—and the blank page problem disappears.