Key Takeaways
- Building trust into a data product from the start is what separates a data product from a dataset, and what determines whether it can support AI systems at scale.
- AI-ready data products require machine-readable semantics, standardized reference data, and governance that extends to the agents and models consuming them, not just the humans.
- Organizations that advance their data product strategy fastest are the ones running foundation-building and business-value delivery in parallel, not waiting for perfect before shipping.
Most organizations have invested heavily in modern data infrastructure. They’ve built data lakes, data warehouses, and data catalogs, and funded cloud migrations and tooling refreshes. And yet, when you ask the people who most need data — like the marketing team, underwriters, and sales directors — whether they can find what they need and trust what they find, the answer is frequently no.
That disconnect was the starting point for a panel discussion I recently moderated, hosted by CDO Magazine and joined by four data leaders who are actively building the organizations and partnerships needed to close that gap:
- Adam Reiser | Sales Director for North America, Huwise
- Arlind Mucaj | AVP, CDO Exchange & Partnerships, Verisk
- Ben Flusberg | Chief Data Officer, Cox Automotive Inc.
- Jessica Jolly | Global Head, Data Office, Alexion Pharmaceuticals, Inc.
Over the course of the conversation, a clear picture emerged: the organizations getting the most out of AI-ready data products are making trust a design principle, not an outcome. It’s not simply about having the biggest platform.
What follows are the themes I think every data leader should sit with, and the ones that will matter most as agentic AI raises the requirements even further.
Why are business users still struggling to find and trust data, despite all the platform investment?
Adam put it simply: your business users don’t have a place they understand to get data from, in a manner that’s understandable. Technical teams invest years in Snowflake, Databricks, and enterprise data platforms that are powerful but opaque to non-technical consumers. The data exists, but it’s inaccessible to the people who need to use it.
Ben expanded on this from Cox Automotive’s own experience. Discovery, he explained, is the solvable part. After establishing an enterprise data marketplace years ago, the organization has reached a point where people know where to go. Trust, however, is a harder problem, and not primarily a technical one.
“Trusted data needs clear ownership. It needs quality standards, lineage, and stewardship over time,” Ben said. “And gaining agreement on ownership is not necessarily a technical problem. It’s often more of an organizational or political challenge.”
I flagged something I see consistently in client engagements at Precisely: organizations treat change management as a downstream concern. They focus on the technology and assume culture will follow. It won’t. When you’re fundamentally shifting how an organization manages its data, you’re also asking teams to change how they do their jobs, and that takes intentional, sustained investment.
The foundation of a data product strategy is an organizational choice. Data marketplaces and discovery tools enable discoverability, but they’re only as useful as the trust infrastructure behind them. Consumability and trust must be built together, or the marketplace becomes another place people search without confidence.
Making the shift from publishing datasets to a data product mindset
This is a topic I asked Jessica to address. She leads the Alexion Global Data Office and has been driving the transition from fragmented, system-generated data to intentional, AI-ready data products across a complex pharmaceutical organization.
“When I think about data as a product,” Jessica said, “that to me signals we need to stop treating our data as a byproduct of our systems and start managing it intentionally so that it’s available for those various personas for consumption.”
A real data product, in her framing, has a clear owner, defined user personas, explicit quality expectations, access rules, and — critically — a reuse mindset. The FAIR principles (findable, accessible, interoperable, reusable) are necessary, but not sufficient. The trust component is what makes them actionable.
Arlind extended this with a sharper framing: a data product is a “reliable promise for a defined customer and outcome.” The product team no longer asks whether a dataset was published. It asks who is using it, what decision it improves, what the data contract promises, and what happens when the underlying source changes.
Adam offered the most intuitive analogy: think of a data product the way you think about ordering something from an e-commerce site. When a package arrives, it comes with everything you need to use it: instructions, packaging, context. A raw database table handed to a non-technical team member is the equivalent of a box with no label and no instructions. Metadata, documentation, and usability features are what make it a product.
What struck me most is that the product mindset shifts accountability. When data is a product, someone owns not just the storage location, but the quality, freshness, and user experience. That accountability is what enables reuse, and reuse is what converts platform investment into business value.
What does it take for a data product to be AI-ready?
I posed this question to Ben, who laid it out in three clear characteristics:
- Data needs to be machine readable, in addition to human readable. That means embedding semantic descriptions: the translation layer between a natural language request from an AI agent and a reliable SQL query or API call. Without that layer, you’re asking the AI to figure it out, which introduces inconsistency and hallucinations.
- Standardization is a force multiplier. The more canonical and consistent your reference data, the less the AI system has to reconcile on its own. That reconciliation work is where errors compound.
- “Governance needs to be extended into this model of AI consuming the data. That involves tracking identities of the agents, setting policies around what agents can access and what they can do with it.”
Adam added a practical dimension: data products need to be accessible by an MCP (Model Context Protocol) server. If AI systems can’t reach the data through a governed access layer, they’ll find another way, and that’s where control breaks down.
Arlind reinforced the compliance stakes here. “If you have made a reasonable attempt to govern the data, to limit what a model can get to, you’ll be given the benefit of the doubt,” he said, referencing regulatory expectations in the EU and elsewhere. “If you just turn your model loose, you’re going to have a problem.”
Jessica tied it back to design philosophy: “If governance is outside the product, we all perceive it as friction. But if we actually engineer it in, we’re embedding the governance by design. And that’s what starts to infuse confidence with our consumers.”
That engineering-in principle is exactly what Precisely and Huwise address together: the Precisely Data Integrity Suite provides the governance, quality, and lineage layer, while the Huwise data marketplace delivers the discoverability and access experience that makes those AI-ready data products usable across teams, partners, and ecosystems. The combination ensures trust is verifiable at the point of consumption.
What should CDOs prioritize to advance their data product strategy through the rest of 2026 and into 2027?
I asked the panel to look forward, and the advice converged on a few themes worth highlighting.
Ben’s core recommendation was one I find myself echoing constantly: drive business value and build the foundation in parallel. Don’t wait for perfect. “If you just focus on the foundations,” he warned, “it’ll feel like a waterfall project that never ends. No one will see the value and they won’t want to fund you.” Find the quick wins, prove ROI, then reinvest that credibility into the harder infrastructure work.
Arlind’s guidance: “Build identity once, define meaning once, and operationalize trust once. Then let many products and AI workflows compound that value.” The organizations that scale fastest aren’t running a hundred AI pilots, but instead making a few foundational investments repeatable.
Jessica extended this with a point I think is easy to lose in the current AI conversation: AI-ready data products aren’t only for AI. “We need to shift from data products as assets to data products as part of our intelligence layer,” she said. The winners will be organizations that connect data, context, quality, access, privacy, and AI governance into one engineered flow — not because agentic AI is the end goal, but because that’s what it takes to serve any consumer of data well.
Adam pointed to a practical risk for the near term: as every platform adds AI features, organizations risk accumulating redundant systems with overlapping capabilities. A single source of truth for data discovery goes beyond being just a governance matter, but an architecture discipline.
From my own vantage point at Precisely, the common thread I see in organizations that are gaining ground is that they’ve stopped treating governance as external to the product. When governance is engineered in from the start — as a design constraint, not a compliance review — it stops feeling like friction and starts functioning as infrastructure. That’s when Agentic AI becomes not just possible, but reliable.
Trusted, reusable, and ready: the standard every AI-ready data product has to meet
Successful data products don’t happen by accident. The organizations getting the most out of their data and AI investments are building trust directly into their data through governance, quality, transparency, and continuous stewardship. That’s what makes it possible for people and AI systems to discover, understand, and confidently use the information they need.
What this panel made clear is that the path forward requires a design shift: from data as a byproduct of systems to data as a governed, reusable asset built for consumption; from governance as a compliance gate to governance as infrastructure; from AI readiness as a future goal to AI readiness as a current design requirement.
For a practical framework on how to get there, I’d encourage you to read Data Product Marketplaces: A Practical Guide for Data Leaders — which underpins much of what our experts discussed in this session. It’s a useful starting point for continuing the dialogue within your organization around building trusted, reusable data products for AI and analytics.
