Eyes on Data: Data Products – The Backbone of AI Optimization

Eyes on Data is a TDAN column published every quarter.

In the race to operationalize and optimize artificial intelligence, most organizations are still asking the wrong first question. They begin with: What can AI do for us?

But leading organizations are reframing the problem: What trusted data do we need to make AI work reliably, repeatedly, and at scale?

This shift is strategic and decisive. AI models might be capturing a lot of the attention, but it is data products that determine outcomes – and drive the anticipated benefits.

If the first wave of AI adoption has taught us anything, it is this: Experimentation is easy; production is hard. Moving from isolated successes to enterprise-wide impact requires more than models and infrastructure. It requires a disciplined approach to how data is created, managed, and delivered. That discipline comes from data products.

Our previous column discussed the journey for successful AI optimization, including trusted data, ontologies, knowledge graphs and data products, ultimately striving for “gold in, gold out.” In this column, we’ll discuss more deeply the critical role of data products as key assets for any AI program.

Beyond Pipelines: The Evolution to Data Products

For years, organizations have invested heavily in data pipelines, lakes, warehouses, and integration platforms. These investments were necessary but were not sufficient. They moved data – but they did not make data usable.

Data products represent a fundamental evolution. They treat data not as a byproduct of systems, but as purpose-built assets designed for consumption. A data product is discoverable, understandable, trustworthy, and reusable. It has defined ownership, clear semantics, and governed access. The DPROD standard formalizes what constitutes a data product including how it is optimized for reuse, intelligibility, and control. Most importantly, a data product is designed with the consumer in mind – whether that consumer is a human business analyst, an application, an operational system, or an AI model.

This product-oriented mindset is what enables scale. Without it, AI initiatives remain trapped in cycles of bespoke data preparation, where each new use case requires rebuilding context from scratch.

Why AI Demands Data Products

AI systems – particularly generative AI – are uniquely sensitive to ambiguity. They do not inherently “know” what data means. They infer meaning from patterns, correlations, and context. When that context is inconsistent, incomplete, or opaque, outputs degrade quickly.

This is why traditional data environments struggle under AI workloads. Data may be technically accessible, but it is rarely AI-ready.

Data products close this gap by embedding three critical attributes directly into the data layer:

• Context: What does this data represent? How should it be interpreted?
• Quality: Can it be trusted for decision-making and automation?
• Control: Who can use it, and under what conditions?

Without these attributes, AI remains based not on certainties but more on probabilities – guessing where it should be reasoning.

With these attributes, AI becomes significantly more deterministic, explainable, and valuable.

The Anatomy of an AI-Ready Data Product

Not all data products are created equal. To support AI optimization, data products must go beyond basic packaging and governance. They must be engineered for machine interpretability as well as human usability.

At a minimum, an AI-ready data product should include:

1. Explicit Semantics

Data must carry its meaning with it. This is where ontologies (such as FIBO for financial information) and controlled vocabularies play a critical role. By formally defining entities, relationships, and constraints, semantics eliminate ambiguity and ensure that both humans and machines interpret data consistently.

2. Embedded Quality Signals

AI systems need to “know” how much to trust the data they consume. Data products should expose quality metrics, lineage, and validation rules as first-class features, not afterthoughts.

3. Interoperability by Design

AI does not operate in silos. It draws on multiple sources across domains. Data products must be designed to integrate seamlessly, aligning on shared meaning rather than just shared formats.

4. Lifecycle Governance

Data products are not static. They evolve. Governance must ensure that changes are controlled, transparent, and backward-compatible where necessary — especially when AI systems depend on them in production.

5. Consumption-Optimized Interfaces

Whether accessed via APIs, queries, or graph interfaces, data products must be easy to consume. Friction at the point of use directly limits AI adoption.

Together, these characteristics transform data from raw material into AI-grade fuel.

The Power of Ontology-Driven Data Products

As organizations mature their data product strategies, a clear pattern is emerging: the most effective data products are ontology-driven.

Ontologies provide the semantic foundation that data products need to scale across teams, systems, and use cases. When combined with knowledge graphs, they enable a level of interoperability and reasoning that traditional data architectures cannot match.

This combination has profound implications for AI:

• Improved Accuracy: AI models augmented with ontology-driven data products produce more precise and contextually appropriate outputs.
• Reduced Hallucination: By grounding AI in explicit knowledge structures, organizations can significantly reduce the risk of fabricated or misleading responses.
• Enhanced Explainability: Decisions can be traced back through the knowledge graph, providing transparency that is critical for regulated industries.
• Faster Time to Value: Reusable data products eliminate redundant data preparation, accelerating AI deployment.

In effect, ontology-driven data products allow organizations to move from pattern recognition to knowledge-driven intelligence.

From Use Cases to Ecosystems

One of the most underestimated benefits of data products is their ability to compound value over time. Each new data product does not just serve a single use case — it becomes part of a growing ecosystem:

• Data products can be combined to support increasingly complex AI applications
• Improvements in one product propagate to others
• Shared semantics enable cross-domain insights
• Innovation accelerates as teams build on existing assets rather than starting from scratch

This is how organizations move from isolated AI initiatives to AI-enabled enterprises.

Data Products as Economic Assets

As data products mature, their role extends beyond internal optimization. They become exchangeable assets that can be shared, licensed, or monetized. (Read more about this in our previous article, Valuing Data with Data Asset Foundations.)

Data products as assets are the foundation of the emerging data economy in which:

• Organizations publish high-quality, interoperable data products
• Consumers discover and integrate them into their own applications and AI systems
• Value flows not just from data ownership, but from data usability

AI amplifies this dynamic. The better the data product, the better the AI performance — and the greater the demand.

This creates a powerful incentive structure: invest in quality, semantics, and governance, and the market rewards you with adoption and value creation.

Operationalizing the Vision

Despite the promise, many organizations struggle to implement data products effectively. The challenge is not conceptual — it is operational.

Success requires alignment across multiple dimensions:

• Organizational: Clear ownership and accountability for data products
• Technical: Architectures that support semantic modeling, graph structures, and scalable delivery
• Cultural: A shift from project-based thinking to product-based thinking
• Standards: Common frameworks that ensure consistency and interoperability

This is where industry standards and communities play a critical role. The EDM Association’s DCAM (Data Management Capability Assessment Model) defines the best practices for building, measuring and managing AI-ready data management programs. The CDMC framework defines the automated control environment that serves as a seamless fabric for data products. And the DPROD (Data Products Ontology) standard ensures reusability and onward controls for safe and efficient data consumption. Used together, these EDM Association standards and best practices reduce ambiguity, accelerate adoption, and create a shared foundation for innovation.

The Path Forward

The journey to AI optimization is, at its core, a journey to better data. Not just more data, but better structured, better understood, and better managed data.

Data products are the mechanism for achieving this at scale. But not just any data products. The future belongs to data products that are:

• Ontology-driven
• Interoperable
• Governed as products
• Designed for both human and machine consumption

Organizations that embrace this model will find themselves with a compounding advantage. Their AI systems will perform better. Their time to market will shrink. Their ability to innovate will expand.

Those that do not will remain stuck in cycles of reinvention, where each AI initiative begins with the same foundational problems.

A Call to Build, Not Just Consume

It is tempting to view AI as something to consume and implement — a capability delivered by vendors, platforms, and models. But the real differentiator lies in what organizations build themselves: their data foundation.

Data products are not just enablers of AI. They are the infrastructure of intelligence.

The organizations that lead in the next decade will not simply adopt AI. They will engineer the data ecosystems that make AI reliable, scalable, and transformative.

The question is no longer whether to invest in AI. It is whether you are building the data products that will allow AI to deliver on its promise.

Learn More

• Get involved with the EDM Association’s Data Products & Marketplace Forum to collaborate on best practices, use cases and more
• Learn more about our collaboration with the Isle of Man to develop Data Asset Foundations, the world’s first statutory framework to define data as an asset
  • Read more in our previous article, and contact us to get involved, whether you are a data-rich company or a corporate service provider
• Explore EDM Association best practice frameworks, including DCAM and CDMC
  • Please note: DCAM is available exclusively to EDM Association member organizations. Not yet an EDM Association member? Learn about the benefits of membership or contact the team.

This quarter’s column contributed by: 

Jim Halcomb, Chief Research & Development Officer, EDM Association

Jim Halcomb is a strategy, data management, and cybersecurity executive with 30 years of international business experience. Jim leads EDM Association’s Communities of Practice, Best Practices Frameworks (DCAM & CDMC) and Training & Certification programs.