What’s the most common constraint that consistently determines whether a company can move from basic AI use cases to advanced AI interoperability? Data governance.

While most companies understand that AI governance introduces new dimensions – including model behavior, fairness, explainability, and decision accountability – all of these rely on the integrity, structure, and usability of the underlying data.

Data governance asks: is this data accurate and properly owned?

AI governance asks a different question: is this system’s behavior fair, traceable, and stable over time?

One cannot substitute for the other, but neither can succeed without the other. The gap between those two questions explains why so many companies move quickly into pilots but struggle to operationalize AI across meaningful, enterprise-wide use cases.

Traditional data governance creates the illusion of readiness

Common issues with data governance stem from the fact these programs rarely translate into better data in practice, and AI makes this more visible.

These models do not consume policy documents or rely on institutional knowledge. They need data that is consistent, contextualized, and immediately usable.

Therefore, a model consuming ambiguous or inconsistently defined data does not add interpretive judgment; it learns the ambiguity and encodes it at scale.

Many organizations have spent years building data governance programs that emphasize ownership models, policy frameworks, and documentation standards.

These programs often produce catalogs, glossaries, and control structures that signal maturity and create confidence at the executive level, without actually creating better, more usable data.

Traditional approaches tend to focus on defining data rather than operationalizing it. Data owners are assigned, policies are written, and critical data elements are documented, but the output often lives outside the systems where data is created and used.

Over time, this becomes a static layer of governance that satisfies compliance requirements without materially improving how data flows through the organization or how it supports decision-making.

AI requires data that encodes human knowledge, not just data as is

For AI systems to operate accurately and at scale, data needs to be enriched with meaning.

This includes:

• Clear definitions
• Lineage
• Transformation logic
• Valid value constraints
• Business context explaining why the data exists and usage instructions

The data must also carry the context and tacit knowledge that humans would otherwise supply through interpretation.

The goal is not to remove context from data but to embed it so completely that the AI system does not need a human intermediary to understand what the data means.

If a user needs institutional knowledge to correctly interpret a field or metric, an AI system will face the same limitation and will have no way to signal that it is struggling.

Quality cannot be an abstract concept defined in a policy document. It must be measurable, enforced, and visible within the data itself.

This changes the role of governance from oversight to execution. Governance becomes something that is built into pipelines, embedded into data models, and continuously validated as data moves through the system.

Without that level of structure, AI systems are forced to interpret incomplete or ambiguous information, which leads to unreliable outputs and limits the ability to trust or scale those systems.

Poor data governance practice will stand in the way of advanced AI interoperability

Scaling beyond AI within narrow, well-defined use cases is where most efforts stall. The underlying data is not structured in a way that allows it to be reused or combined effectively, and governance practices are not designed to support that level of interoperability.

The failure mode is predictable: a model trained across business units inherits conflicting definitions of a core entity, such as an active policy, a qualified lead, or an in-force risk, and learns those inconsistencies as signals.

The outputs are confident, fluent, and wrong. No amount of model tuning corrects for a governance failure that was baked in before training began.

At Hippo, avoiding that outcome required treating the governance foundation as a prerequisite, not a follow-on.

Our team set out to build AI-driven analytics that could answer not just “what is our average roof age across policies in force,” which is a question any BI tool can handle.

We needed a system that could handle more consequential questions that require AI to reason across property attributes, underwriting guidelines, and policy data simultaneously.

Getting there required treating semantic metadata and data context as first-class engineering concerns, not documentation afterthoughts. The data governance foundation had to come first; the AI governance and capability followed from it.

This alignment is critical. Without it, organizations remain limited to incremental improvements rather than achieving the kind of transformation that AI promises.

What needs to change to support AI at scale

Closing the gap between data governance and AI governance requires a fundamental shift in priorities.

Organizations need to move away from viewing governance as a documentation exercise and toward treating it as a core component of their data architecture, helping to reduce AI governance risk.

To put this into practice, organizations should:

• Invest in data platforms that embed context and quality directly into the data
• Enforce data contracts at pipeline boundaries, carrying semantic metadata through to the systems that consume it
• Ensure that lineage is traceable from model output back to source
• Align governance practices with how data is actually used, not theoretically defined
• Recognize that governance is not a one-time initiative, but an ongoing capability that requires continuous attention and adaptation.
• Evolve roles of data owners and stewards from maintainers of documentation to active curators of data

For digitally native companies, this shift may already be underway, driven by the need to move quickly and deliver measurable outcomes.

For more traditional organizations, the challenge is greater, but the path forward is becoming increasingly clear, and will define their success in deploying useful AI systems that support enterprise objectives.

About the Author:

Robin Gordon is Chief Data Officer at Hippo. Gordon has spent the past decade in executive roles as Chief Data Officer and Chief Technology Officer, driving growth, digital transformation, and AI adoption at global companies including Blackstone, KPMG, Metlife and CoreLogic.