I’ve been having more and more conversations with data leaders and architects who are all wrestling with the same challenge: How do we make AI actually work in the enterprise?
It’s no longer about building chatbots or plugging in a model. The real question is, how do we give AI the context it needs to reason, decide, and act the way our business does?
And that’s where a concept we’ve known for decades has come roaring back into relevance: data modeling.
AI Doesn’t Just Need Data. It Needs Context
For years, we treated data modeling as a technical, behind-the-scenes discipline, something done to make BI dashboards run faster or reporting easier.
And it worked because back then, we weren’t asking machines to make decisions.
But now, with the rise of Agentic AI systems that can observe, reason, and act autonomously, the problem has changed.
AI can read data. But it can’t understand your business logic unless you model it.
Let’s make it concrete:
- A customer becomes an account only when it’s linked to an invoice.
- A lead becomes a qualified opportunity when they attend a product webinar.
- A candidate becomes an employee when there’s a signed offer.
There’s no universal rule for any of these. Each business defines them differently.
That’s exactly what AI struggles with – understanding meaning, relationships, and transitions.
And that’s what data modeling was always meant to capture.
From Reporting to Reasoning: The Evolution of Data Modeling
Ralph Kimball – one of the founding architects of modern data warehousing, understood this long before AI was on the scene.
His dimensional-modeling methodology emphasized structure with purpose: connecting business processes with data relationships.
In Kimball’s world, facts represented measurable events (like sales or transactions), and dimensions provided the business context (like customers, regions, or products).
It wasn’t just about tables but it was about modeling how the business actually thinks.
Today, that thinking is more relevant than ever.
AI and automation depend on semantic understanding – the relationships between entities and the business rules that govern them.
Kimball’s approach still provides that blueprint:
- Conceptual models capture meaning (“What is a customer? What is a transaction?”).
- Logical models define relationships (“How are customers connected to products or invoices?”).
- Physical models operationalize those relationships in data systems.
When AI understands these models, it stops guessing and starts reasoning.
Example: Turning Data into Context with Automation
One of our enterprise customers, a global financial-services company, came to us with a familiar challenge:
Their AI tools could analyze customer data, but they couldn’t interpret it.
Each department had its own definitions – “client,” “account,” “relationship,” “portfolio” and none of those meanings matched across systems.
Their AI models would recommend actions based on incomplete or conflicting data.
We helped them build an event-driven, modeled data layer using Workato’s platform, mapping business definitions to event structures in real time.
For example:
- When a “portfolio value change” event was triggered in their CRM, it was automatically linked to “customer risk category” data from the core system.
- Our automation recipes enriched and standardized those attributes across downstream processes, ensuring the AI always operated within the same contextual framework.
The result?
AI-driven decisions that were consistent, explainable, and aligned with business rules.
In other words, AI that finally understood the business.
Example: From Chaos to Clarity in Retail Data
Another customer was a large global retailer, faced a similar issue but in a different form.
Their marketing AI models were brilliant at predicting customer churn, but not very good at understanding why it was happening.
After reviewing their data, the problem wasn’t quantity, it was context.
Customer, order, and product data were spread across dozens of SaaS systems, each defining “return,” “cancellation,” or “promotion” differently.
By applying a Kimball-inspired conceptual model, they redefined their core dimensions, aligning customer behavior data with operational metrics through modeled structures.
Once the business logic was modeled and automated across systems, their AI agents began identifying patterns that previously looked random.
They discovered that “returns” following specific promo codes weren’t quality issues at all, they were logistics delays masked as customer dissatisfaction.
That’s the power of context. The data didn’t change. The model did.
The Missing Link: Context in the AI Era
Recent announcements from Atlan further validate this shift. At the company’s Activate 2025 event, Atlan positioned context as the missing foundation for trustworthy AI.
Their new App Framework, Metadata Lakehouse, AI Governance Studio, and Data Quality Studio underpin the idea that models may be easy, but missing context is what breaks most enterprise AI initiatives.
One key insight from Atlan:
“The models turned out to be easy. The data and context are what’s breaking us.”
That echoes exactly what we’re seeing in our engagements.
AI projects stall not because the algorithms lack power but because they lack shared meaning, governed relationships, and consistent context.
From Kimball to Cognitive: The Data Model as AI’s Brain
In the era of BI, data models powered reports.
In the era of AI, data models power reasoning.
Modern AI doesn’t just need rows and columns, it needs ontologies, semantics, and logic that define how data behaves in context.
This is why every successful enterprise AI implementation starts with conceptual modeling, not code.
It’s how AI systems know that:
- “Closed Won” is a signal to trigger a billing workflow.
- “Onboarding complete” means a support ticket should be created.
- “Contract expires” implies an upsell or renewal opportunity.
Without these modeled relationships, AI is working blind.
How Our Platform Fits In: From Modeling to Operational Intelligence
At Workato, we see modeling not as a theoretical exercise, but as a living framework for how data flows and decisions happen across the enterprise.
We help enterprises build AI-ready architectures with three core capabilities:
- Event Streams: Handling semi-structured data (JSON, APIs, webhooks) and transforming it into modeled, real-time business events. This makes it possible for AI and automations to “understand” state changes as they happen.
- AI@Work: Applying reasoning and enrichment models that interpret event data using business context thus turning an event into insight or action.
- Intelligent Orchestration: Connecting every modeled entity – customer, order, asset, or transaction into cross-functional workflows that think and act coherently.
Together, these capabilities form the nervous system of the Agentic Enterprise, where every decision, every action, and every insight operates on clean, contextual, and modeled data.
The Bottom Line: AI Thinks in Models
Ralph Kimball once said that the goal of modeling is to “bring the business and the data together.”
That’s exactly what we need to do again – this time, for AI.
Because AI doesn’t just feed on data; it thinks in relationships, hierarchies, and meaning.
That’s what a data model provides – a framework for understanding.
The companies winning with AI today aren’t just feeding models more data. They’re feeding them better context – shaped by the same modeling discipline that powered the data-warehouse revolution decades ago.Data modeling isn’t back because it’s nostalgic.
It’s back because AI can’t work without it.