Semantic Model — and How this Helped in the Tableau Hackathon

Intro

When people hear the words semantic model, it can sound big, technical, and confusing.

But the idea is actually simple.

A semantic model is just a smart layer that helps a system understand what your data really means. It tells the system what each field is, how numbers should be calculated, how tables connect, what business words mean, and how to answer questions in the right way.

In our hackathon project, the semantic model was one of the most important parts of the solution. It helped turn raw data into something the agent could understand, trust, and use.

Here you can see the assets and the data (data model, relationship, calculations, fields etc) are all stored in the Semantic Model - in simple terms treat it as a Data Source.

Step 1 — What is a Semantic Model?

Think of data like a giant toy box.

Inside the box, there are many pieces:

• customer names

• product values

• revenue numbers

• dates

• statuses

• account types

But if you just throw everything into one box, nobody knows what anything means.

A semantic model is like putting labels on every toy, organizing them into groups, and adding little notes like:

• what this thing is

• how it connects to another thing

• when it should be used

• how to calculate it

• what words people might use when asking about it

So instead of showing the AI or the user a messy pile of data, the semantic model gives them a clean and organized map.

Step 2 — Why It Mattered in the Hackathon

In the hackathon, the goal was not just to show a dashboard.

The goal was to build a solution that could actually understand the business.

That is where the semantic model became powerful.

Instead of forcing the user to know table names, column names, or raw data logic, we created a smarter layer that explained:

• what the main business objects were

• how they were related

• what the important metrics meant

• how the agent should talk about them

• what rules and business language it should follow

This made the solution feel less like a report and more like a business assistant.

Step 3 — The Simple Building Blocks

A semantic model is usually built from a few basic parts.

Fields

Fields are the small pieces of information inside the data.

Examples:

• Customer Name

• Revenue

• Stage

• Created Date

• Region

• Product

• Opportunity Amount

A field is just one fact.

You can explain it like this:

A field is like a label on one box.

Calculations

Calculations are little math or logic rules.

Examples:

• total revenue

• conversion rate

• average order value

• win rate

• growth percentage

A calculation helps turn raw data into something more useful.

You can explain it like this:

A calculation is like telling the system how to count or compare things properly.

Relationships

Relationships show how different groups of data connect.

For example:

• a customer can have many orders

• an account can have many opportunities

• a product can belong to a category

• a region can contain many customers

You can explain it like this:

Relationships are like roads between houses on a map.

Without them, the system knows the houses exist, but it does not know how to travel between them.

Business Names and Descriptions

This is where the model becomes human-friendly.

Instead of only showing technical names, we add better explanations:

• what the field means

• when to use it

• what the business calls it

• what the user might ask for

This helps the AI speak in business language instead of machine language.

Step 4 — What We Added to Make It Smarter

This is where the semantic model became really useful for the hackathon.

We did not just list fields.

We also added meaning.

That included:

• clear field names

• better descriptions

• useful calculations

• logical relationships

• business-friendly wording

• important guidance for the agent

This helped the agent understand not only what the data is, but also how the business thinks about the data.

Step 5 — Business Preferences Explained Simply

One of the most useful parts was the business preferences.

This sounds fancy, but it is actually simple.

Business preferences are like giving the agent a little rulebook.

We tell it things like:

• which metrics matter most

• what the business calls different things

• which terms should be used

• what to avoid

• what logic should come first

• how to answer in a way that matches the company’s style

Here's an explanation:

If the semantic model serves as the map, business preferences act as the driving guidelines.

They assist the agent in keeping on the correct path.

Without business preferences, the AI might still provide answers, but they could be unclear, generic, or inconsistent.

Incorporating business preferences makes the AI significantly more effective.

Step 6 — Teaching the Agent

The agent does not magically understand the business.

It needs help.

That help comes from:

• the semantic model

• the field definitions

• the calculations

• the relationships

• the business preferences

• the naming rules

• the descriptions

In simple words:

We were teaching the agent how to think about the data.

We were not just giving it numbers.We were giving it context.

This is the part many people miss.

A good agent is not only about prompts.A good agent needs a good foundation.

And in this case, that foundation was the semantic model.

Step 7 — The Last-Minute Updates

One of the most important parts of the hackathon was that we kept improving the semantic model until the end.

At the last minute, we updated and refined the model so the solution would be clearer, more useful, and better aligned with the final experience we wanted to show.

That included improving:

• descriptions

• business wording

• logic

• structure

• agent guidance

This was important because small improvements in the semantic model can make a big difference in how intelligent the final solution feels.

Sometimes the dashboard gets the attention.

But behind the scenes, the semantic model is often doing the heavy lifting.

Step 8 — What I Learned

The biggest lesson was this:

AI is only as smart as the business meaning behind the data.

You can have a beautiful dashboard. You can have a clever agent. You can have strong technology.

But if the data has no clear structure, no trusted logic, and no business meaning, the result will still feel weak.

That is why semantic models matter.

They help turn raw data into something understandable, useful, and ready for both people and AI.

What I like the most is that you can see the Trailhead courses and learning notifications:

Closing Thoughts

In my view, the semantic model was more than just a technical setup during the hackathon.

It was a key factor in the solution's success.

It facilitated the connection between the business, the data, the agent, and the user experience.

This is why I am convinced that semantic models will become crucial in modern analytics, particularly in the realms of Tableau Next, self-serve BI, and AI-driven decision-making.