Unlocking Agile Analytics with Composable Semantic Models from AtScale

In today’s fast-paced world, companies must utilize data to make informed decisions. However, with so much data coming from various sources, company records, sales systems, cloud storage, and live data feeds, it’s challenging to maintain everything in a neat, consistent, and helpful manner. Imagine trying to sort thousands of LEGO pieces from different sets, all mixed together in various boxes across multiple rooms! This is what it’s like for businesses trying to get a clear and reliable picture of their performance.

That’s where composable semantic modeling comes in.

AtScale’s Composite Modeling capabilities let data teams build modular, reusable semantic models that can be combined to deliver a single source of truth, without duplicating logic or replicating data. This approach enables agility and consistency at scale.

What Is Composable Semantic Modeling?

Composite Modeling is a game changer for how data teams work, as it enables them to combine important information from multiple data platforms, including big cloud storage services such as Snowflake, Databricks, Google BigQuery, Amazon Redshift, and even legacy company databases. Teams can contribute their own Models, which contain all the relevant metrics, dimensions, and calculations, and provide them as a single source of truth. Then, other teams can take those models and bring them together to create a unified view across multiple areas, without the overhead of copying that logic. 

Why Composite Modeling Matters

Here are some significant benefits of Composite Modeling, showing how it improves how companies use data:

• Use data from different teams without copying it: Let your teams build and publish models that represent the correct data from their operations, allowing others to utilize those models and providing a single source of truth for each critical metric.
  
• Build models that are easy to reuse and update: By having a single model that is owned by a single team, they can control how that data is represented and leveraged. Now, with composite modeling, you can pull that model into your own to ensure that the data will always be correct. When a change is made to the lower-level model, the composite model inherits it automatically without requiring user intervention.
  
• Give analysts more freedom, without losing control: Let your teams build the reports they want and need using the tools of their own choice. With a composite model feeding the data, you know you can trust the results. Let your experts build their own models and let your teams pull those together to tell the complete story.

Real-World Example: Decomposing the TPC-DS Benchmark Model

To show this in action, let’s walk through how we use Composite Modeling with our own TPC-DS benchmark dataset.

The benchmark model we provide for our tutorials is a beautiful but LARGE model with a LOT of complexity. It’s solving significant hurdles like joins and relationships, but it’s a lot to grasp.

Initially, this was a single, complex model with dozens of tables, relationships, and calculations—effective, but hard to maintain and scale. Using composite modeling, we broke it into smaller, domain-specific models:

• Catalog Sales
• Inventory
• Purchases by Channel
• Store Promotions
• Store Returns
• Store Sales
• Web Sales

Each of these became its own AtScale semantic model, complete with owned dimensions, measures, and calculations. These were then published to a shared repository, where other teams could discover and reuse them.

How It Works in AtScale

To create a unified view, we built a Composite Model in AtScale’s Design Center by dragging these component models onto a canvas. AtScale automatically recognized shared dimensions and relationships, allowing the composite model to retain all the analytical power of the original, without the monolithic sprawl.

The result? A simpler, more modular, and more maintainable semantic layer that anyone across the business can trust and use.

We’ll create an AtScale model for each of these as we would any other model and save them to our common repo for distribution, collaboration, and use.

There is nothing special we have done when building these smaller models compared to our original TPC-DS benchmark model; we simply moved each functional space to its own model, with its own measures, dimensions, and calculations.

To provide a single view of data for these six separate models, we’ll create a Composite Model. We do this in AtScale by adding a new Composite model in Design Center, which lands us on a blank canvas. 

Now we simply need to drag and drop those smaller models onto the canvas, and our composite model will begin to take shape.

And once we’re done, we have a single Composite Model to deploy, which contains all the same measures, dimensions, relationships, and calculations our mega-model had, but in a much easier-to-consume and understand way.

Why Composable Modeling Is the Future of the Semantic Layer

Composable modeling aligns with broader trends in the modern data stack:

• Data Mesh and Domain Ownership: Allow teams to own and publish trusted data models.
• LLM-Ready Architecture: Feed AI agents consistent, explainable semantics.
• BI Consistency Across Tools: Guarantee identical answers across Excel, Tableau, Power BI, and more.
• Cloud-Scale Performance: Query live against cloud platforms like Snowflake, Databricks, BigQuery, and Redshift—without data movement.
  

This architecture supports not just analytics and dashboards, but also natural language query (NLQ) and agentic AI use cases. It gives enterprises a scalable, governed foundation for data-driven decision-making.

The Composable Semantic Layer, Operationalized

AtScale’s composite modeling capabilities enable the scaling of governed, reusable business logic across teams, platforms, and tools. It provides data engineers with the flexibility to manage complexity, business analysts with the confidence to explore data freely, and executives with the insights they need to make high-impact decisions.

As Gartner notes in their 2024 Market Guide:

“Composable data and analytics supports a more agile, modular, and federated approach to data consumption and semantic governance.”

With AtScale, composable semantic models are not just a concept; they’re an operational reality.