The Hidden Problem Behind AI for Data
Building AI agents that can accurately query business data isn't just a technical challenge—it's a knowledge management problem. For our agents to work reliably, they need to understand your business context. How does your company define "churn"? What constitutes an "active customer"? When you say "quarterly revenue," which fiscal quarters are you referring to?
These definitions form the foundation of how your business operates. Yet in most organizations, this critical knowledge is scattered, inconsistent, and surprisingly hard to capture.
The Traditional Approach: Top-Down and Broken
The conventional solution has been to hire data stewards who act as bridges between business teams and data teams. Their job is to:
- Interview business stakeholders across departments
- Collect and document how each team defines key metrics
- Create centralized definitions that everyone can use
- Maintain these definitions as the business evolves
This top-down approach makes sense in theory. In practice, it's often a nightmare.
The process is linear and unscalable. A data steward can only interview so many people, attend so many meetings, and document so many definitions. As organizations grow and change, keeping up becomes impossible.
Knowledge stays siloed. Even with the best intentions, critical business knowledge remains locked in people's heads. The sales team knows how they calculate customer lifetime value, but that knowledge might never make it to the official data dictionary.
Definitions become stale. Business evolves faster than documentation. By the time a definition makes it through the approval process, the business reality might have already changed.
A Better Way: Bottom-Up Data Governance
While building AI agents at Wobby, we stumbled onto what we believe is a better approach: bottom-up data governance.
Here's how it works in practice:
Imagine a business user asks our AI agent: "Can you tell me about customer churn in the last quarter?"
If the agent doesn't have a clear definition of "churn" for this company, instead of making assumptions or failing, it asks a follow-up question: "How do you define churn? Is it customers who haven't purchased in 90 days, or customers who've formally canceled their subscription?"
The user provides their definition: "Churn means any customer who hasn't made a purchase in the last 6 months."
Now something powerful happens:
- Knowledge is captured - The agent can now run the query using this definition
- Context is preserved - This definition is saved and tagged with who provided it and when
- Governance remains controlled - The data team can review, approve, or refine this definition
- Learning is distributed - The next person who asks about churn benefits from this knowledge
-
Why This Changes Everything
This approach flips the traditional model:
Instead of interrupting workflows to gather definitions, definitions emerge from actual work. Business users aren't pulled into separate meetings to define metrics—they're defining them as part of getting their questions answered.
Knowledge capture becomes continuous, not periodic. Every interaction is an opportunity to learn something new about how the business really operates.
Definitions stay fresh. When business logic changes, it's captured immediately in the next query, not months later in the next governance review.
The data team stays in control. While definitions are captured bottom-up, approval and standardization still flow through proper governance channels.
The Bigger Picture
What we've discovered goes beyond making AI agents smarter. Bottom-up data governance represents a fundamental shift in how organizations can manage their most valuable asset: institutional knowledge.
Data governance becomes embedded in the daily flow of work. Business users contribute their expertise naturally, data teams maintain control and consistency, and everyone benefits from a richer, more accurate understanding of what the data actually means.
This doesn't eliminate the need for data stewards or formal governance processes. Instead, it makes them more effective by ensuring they're working with fresh, real-world knowledge rather than trying to extract information through artificial processes.
Looking Forward
We're still in the early days of this approach, but the potential is clear. When AI agents can capture business knowledge as part of their normal operation, we create a virtuous cycle: better definitions lead to more accurate insights, which build trust, which encourages more usage, which captures even more knowledge.
The result is better organizational learning. And in a world where business moves faster than traditional governance can keep up, that might be exactly what we need.
At Wobby, we're building AI agents that don't just query your data—they learn your business. If you're interested in exploring how bottom-up data governance could work in your organization, we'd love to hear from you.
