More than datasets: How data products make data usable for humans and AI

In an ideal world, data should function like a utility; reliable, invisible, and essential. Just like the electricity that powers an office or the water that runs through a building, you shouldn’t have to think about how it works; you should only need to know that it’s there when you turn the tap.

However, we don’t live in that ideal world yet. For the vast majority of organizations, the "data faucet" is still leaking, the pipes are often disconnected, and the quality of what’s coming out is inconsistent. While a handful of global leaders might treat data management as a "solved problem," for most, it remains the primary bottleneck to operational efficiency.

My work has always been about occupying that service-oriented space: acting as the enabler who ensures the infrastructure is reliable enough to support the weight of the business. It’s a role that focuses on the "unseen" work; the governance, the maturity, and the strategic alignment, that allows a company to move from simply "having data" to actually "running on data." And right now, that space is undergoing one of the most profound transformations I have ever witnessed.

The shift from "how" to "what"

For the first decade of my journey in data consulting, I was the technical expert. I was under the hood, building the pipelines and tuning the engines. In the last six plus years, however, my focus has shifted toward the human and organizational side of the equation: data governance, data maturity, and how business units really consume information to drive strategy. No two organizations are identical. I have seen everything from small, agile teams to massive, legacy-heavy enterprises. This perspective has taught me that while technology changes, the fundamental goal remains the same, that is, enabling the business to move.

We are currently living through a massive industry shift. For much of the last two decades, the conversation in our industry was dominated by the technical "how." We focused on the mechanics of construction. While many credit AI and agentic workflows for this, the change actually started long before the current hype. We are finally seeing the emergence of industry-wide standards and agentic productivity tools that most professionals can agree on. These advancements allow us to build with a level of speed and confidence that was previously impossible. But as the technical barriers begin to lower, they reveal a much more intimidating challenge: the "what."

Now that we can build faster, we are forced to ask:

• What should we be building in the first place?
• What assets do we already have in our inventory?
• What data is actually safe and useful for me to consume?

The rise of the data product

To answer these questions, and to bridge the gap between technical output and business value, we need a concept that translates across the entire organization. This is the "data product" opportunity. The beauty (and the frustration) of the term "data product" is that it lacks a universal, rigid definition. This is a good thing. It allows each organization to define its own "productness" (kudos to Nick Zervoudis and Juha Korpela for the term) based on its unique culture and operational needs. However, while the definition is flexible, the mindset required to execute it - data product thinking - is very specific.

However, treating data as a product requires a very specific mindset, to be specific a shift toward "data product thinking." In my experience, a successful data product is a combination of three core elements, but they need to be prioritized correctly:

1. Product management (the priority): This is the most critical element. It’s about understanding the user's needs, the problem being solved, and the value being delivered. Without product management, you just have a dataset; you don’t have a product
2. Metadata (the context): This is the documentation, the lineage, and the "instruction manual" that makes the data discoverable and usable by someone who didn't create it
3. Technology (the tool): Ironically, this should play the smallest part. Technology is just the delivery vehicle. If you lead with tech, you usually end up with a solution looking for a problem.

Defining the boundary

The most significant friction in any organization occurs at the boundaries. With data development, everything is a lot easier when you own the data, the development flow, the dependencies etc. But when you need to use data owned by another team, there’s a lot more friction. In my view, this is exactly where the “data product” thinking will have it’s most significant impact.

A data product is a package of information designed to cross that boundary safely and efficiently. To ensure this relationship works, we can rely on data contracts. These are the formal agreements between the producer and the consumer, specifying exactly how the product is shared, what the quality standards are, and how it should be used. This turns a "favor" between departments into a reliable, governed service.

Semantic clarity for an AI world

While I’ve always been pragmatic about technology preferring to focus on current productivity rather than chasing the "bleeding edge”, it is impossible to ignore the role of AI in this evolution.

For AI agents to be effective, they require more than just access to raw data; they need semantic understanding. They need to know the meaning behind the numbers. Data products, by their very nature, provide this. By wrapping data in metadata and product management principles, we create a map that both humans and AI can follow.

We are moving into an era where the "service" of data is the most valuable asset an organization has. By treating data as a product, we aren't just cleaning up the pipes; we are ensuring that the engine of the business has exactly the fuel it needs to grow. As an enabler of business operations, my goal has always been to make data usable. Whether that user is a human executive making a strategic decision or an AI agent executing a tactical task, the foundation remains the same, namely a well-defined, well-governed data product.

Want to make your data more usable across teams and AI use cases?

Find out where to start with data product thinking - talk to our experts.