Any conversation these days, no matter how far removed from technology it may be, eventually turns to AI. Companies are racing to adopt and build it, whether that means large language models (LLMs), machine learning (ML), or AI features inside the tools they already use.
In that rush, many organizations are making a critical mistake: they're neglecting data modeling. With so much attention on tools that promise instant answers, it's easy to assume traditional engineering practices matter less now. The opposite is true.
Think of AI as a master chef. If you ask them to cook in a kitchen they’ve never seen, with ingredients scattered in unmarked drawers and cabinets, they’ll fail. But give them a well-organized kitchen, and they can create magic.
The same applies to your data warehouse. AI can process huge amounts of information, but it depends on the quality of the data underneath. If that data is inconsistent, duplicated, or poorly defined, the AI won’t have the context it needs to provide accurate answers.
Take a simple question: "How many active customers do we have?" It sounds straightforward, but is an active customer someone who purchased in the last 30 days, or an active subscriber who hasn't bought anything recently?
If different systems use different definitions, you end up with conflicting answers. These aren't just traditional modeling problems; they are AI problems. Without a consistent model, AI will rely on one system's context and answer confidently, even if that answer doesn't align with the business reality.
Whether you use traditional ETL tools like Informatica or modern workflows like dbt, the goal remains the same: creating context. For an LLM, context is the information that explains what the data means, so it can interpret a question and give an accurate answer.
While Informatica is powerful for complex integrations, dbt treats transformations like software development (version controlled, tested, and reused). Ultimately, both tools, along with most other ETL tools, can create the consistent business definitions your AI needs to succeed.
A well-designed model doesn't just organize tables and columns. It creates a cohesive definition of what the data actually means for the entire organization. Business processes also need to align, so there are consistent guidelines for implementing and updating data.
Models need to be scalable and maintainable. Over the years, new systems get integrated and others get retired. A good model will be able to adapt to these changes, while also maintaining consistency in both context and quality. The context is key because, even when data is organized and readable, AI can't fill in the gaps like humans can.
AI only has what you give it access to. While humans can go to others to get answers, AI can only ask a follow-up question or start hallucinating. But, if the model maintains consistency with its context, the AI can keep up with the changes.
In real-world applications, such as a chatbot that answers questions about business data, systems like Snowflake Cortex Analyst rely on a semantic model: the definitions, metrics, and relationships that tell the LLM what each table and column means. If that model lacks context about where each column comes from or how a metric is calculated, the system may query the wrong field because the user's question matches a keyword but not the right business context.
At its core, data modeling gives reporting and AI the structure they rely on. If the model combines different definitions poorly, you are left with multiple answers to the same question. Conforming these systems to a singular definition is exactly what AI relies on to be effective. The foundation of any successful AI product starts with engineering fundamentals. A shaky foundation will lead to a poor product that flat out doesn’t work. If you want to depend on AI, you must look at the groundwork it’s built on.
I like to think of AI as a BI (Business Intelligence) report that can talk back. The same principles apply; reports and dashboards depend on a sound data model and on the context that tells people how to read them. AI needs that same context and complete data to generate that same answer. Data modeling isn’t dead. In the age of AI, it’s becoming indispensable.
Wondering if your data model is ready to support AI? OneSix builds the data foundations that make AI work.
