Notes from a Data Leader:

Making Data Easier for AI to Understand

Data Should Speak for Itself

I've spent most of my career working with business and data leaders as they navigate how organizations use data to make decisions.  

Over time, I've found that the same questions tend to surface across organizations, often long before they become broader industry conversations.

These notes are my attempt to capture a few of those patterns as they unfold.

One thing I've noticed over the past year is that the conversation around data and AI has changed.

A year ago, most discussions were about what AI could do. Today, the questions are much more practical. Why does one AI application perform well while another struggles? Why does every new use case require so much setup? Why are teams spending so much time validating answers before they trust them? And why do costs keep climbing as organizations try to scale?

Those questions are usually treated as separate challenges. The more I think about them, the more I see them as different expressions of the same underlying issue.

We’ve become very good at making data technically accessible. Companies have poured billions into data warehouses and cloud platforms. Every organization has tried or implemented a catalog that makes the discovery and search of data that exists easy; but it doesn’t tell you what data should/can be used and for what.  

AI reminds us that available isn’t the same as understandable.

Before AI, that distinction didn’t matter as much. Business context lived alongside the data, in documentation, governance policies, dashboards, and through the experience of the people who used it. That worked because the people consuming the data already had an understanding of the business.

As AI becomes an independent consumer of data, that model breaks down. Context can no longer exist somewhere else in the organization. It has to be embedded with the data itself.

Where the Missing Context Lives

Imagine asking a simple question: “What was our churn rate last quarter?”

Finding the data is now the eas(ier) part. The hard part is understanding what “churn” means in your company. Which definition applies? Which source is authoritative? Are there exceptions? Has the business definition changed over time? Or if the same definition is used across the organization?

In most organizations, those answers exist. They’re captured across disparate systems: governance policies, semantic models, documentation, dashboards, and often in the heads of business analysts.

The challenge is that much of that business understanding still lives outside the data itself.

Every time AI encounters a new task, it has to reconstruct that understanding before it can begin reasoning. It searches documentation, infers relationships, compares examples, drawing its own conclusions, filling in gaps that experienced analysts would know to stop and question.

That’s where a surprising amount of effort, and cost, comes from.

This isn’t a theoretical problem. Our own research backs this up: 68% of data practitioners say their data isn’t reliable enough for AI, and 87% say more contextual data would directly improve decision speed (The Modern Data Report 2026). These aren’t data scientists complaining about tools, these are the people closest to the problem telling us the foundation isn’t ready.

The barrier isn’t the AI itself, it’s what the AI has to work with.

“AI doesn’t get the benefit of tribal knowledge. Most models don’t browse Slack channels, and agents don’t remember why Metric A is ‘more correct’ than Metric B as discussed on a call two years ago. AI only knows what the system itself makes explicit. Data without context is noise, and AI amplifies that noise.” —Animesh Kumar, Modern Data Report 2026: What 500+ Data Leaders and Experts Say on AI-Readiness

Why Data Products Matter

This is one of the reasons I’ve long believed data should be treated as a product.

A good data product doesn’t just make data accessible. It leverages context and makes business understanding behind the data reusable.

Instead of leaving definitions, governance, quality expectations, and ownership scattered across separate systems, a data product brings them together with the data itself. The next analyst doesn’t have to rediscover what “active customer” means. Neither does the next application. Neither does the next AI agent.

That’s how you stop rebuilding the same foundation every time.

The value goes well beyond AI. When business context becomes part of the data product, every consumer starts from the same understanding. Dashboards become more consistent. Analysts spend less time validating results. New use cases take a lot less effort to build because foundational decisions don’t have to be made again.

AI doesn’t create this discipline. It simply exposes how much organizations have relied on business knowledge that was never directly connected to or captured with the data itself.

Rethinking What Data Teams Produce

I also think this shift changes the relationship between the business and the data team.

Historically, business teams defined how the business worked, and data teams translated those decisions into pipelines, dashboards, and reports. Much of the understanding stayed with the people involved in those conversations.

As AI becomes another consumer of data, business understanding has to become something the organization captures deliberately, not something that’s rediscovered every time a new use case, report, application, or AI agent is built.

That isn’t something data teams can do on their own. Finance owns the definition of net revenue. Sales understands the customer lifecycle. Operations knows the exceptions that never appear in a schema. The opportunity is for data teams to work alongside those experts so those decisions become part of the data itself, instead of remaining scattered across documents and institutional knowledge.

To me, that’s one of the more interesting shifts taking place in our industry right now.

For years, the work of data teams has been measured by how well they collect, integrate, govern, and deliver data. Those responsibilities don’t change. But another one is emerging: helping organizations operationalize business knowledge so it can be reused consistently wherever data is consumed.

That’s a meaningful evolution of the role, and one that extends well beyond AI. It’s also what I’ll keep coming back to in these notes. Next time, I’ll get into what it looks like when data teams make that shift in practice.  

Until next time,

Saurabh