AI for Data Management

I recently had a flashback to my days as a data scientist in the trade show industry.

A simple boardroom question triggered chaos: “How many visitors attended our event?”

Operations came back with one number, pulled from a vendor’s ticketing system. Marketing came back with another, pulled from a BI dashboard. Neither side could fully explain how their numbers were generated.

That’s when I realized: what sounds like a “simple” data question almost never is.

Today, I’ll show you why these kinds of problems are more relevant than ever – and how AI can help solve data management challenges in ways that weren’t possible before.

Let’s dive in!

Why This Old Problem Now Matters More Than Ever

Everyone who has done even a little analytics knows that having different numbers for what looks like the same thing is actually totally normal (and expected). Data definitions reflect different prioritizations.

Operations cared about throughput and efficiency, so their ‘visitor’ definition counted entries first and people second. Marketing cared about conversion, so starting from unique individuals made more sense. Both definitions served legitimate business purposes.

What I learned from the boardroom discussion above is that companies don't actually want unified definitions. They want plausible deniability.

Marketing preferred unique visitors (real people!) because they could benchmark against campaigns. Operations preferred ticket scans because higher throughput made them look efficient. The definitional "confusion" wasn’t confusion at all – it was useful flexibility, depending on who was presenting.

This worked fine when the ambiguity stayed siloed. Marketing used their definition internally, operations used theirs, and the CEO only saw polished slides.

But AI doesn’t respect silos.

Ask an AI assistant “What’s our customer acquisition cost?” and it might pull from three systems with three different customer definitions — and then deliver a precise-sounding answer that’s actually meaningless. AI doesn’t just repeat the old problems, it scales organizational self-deception across the entire company.

Which doesn’t mean AI only makes things worse. Used correctly, it can actually surface and resolve these definitional problems faster than humans ever could.

How AI Actually Helps (3 Fixes)

The solution obviously isn’t to force everyone into a single definition — that’s fighting human nature. Instead, AI can make definitional choices transparent while preserving the legitimate business reasons behind them.

Here are three ways I see this happening:

1) Automated Definition Validation

AI can reverse-engineer how metrics are calculated, explain the logic in plain English, and compare differences side by side.

In my example, it could show that Operations started by counting entry events, while Marketing worked from the other end with unique customer records. Instead of weeks of detective work, you’d see exactly how each calculation was performed — as long as the logic exists somewhere in SQL, code, or another traceable format.

I still remember burning an entire week untangling a 600-line nested SQL query just to figure out what was happening.

I really wish I’d had AI back then.

2) Knowledge Graphs

Think of “Visitor” as an entity in a knowledge graph. Marketing and Operations each attach their own definitions: unique individuals on one side, ticket scans on the other.

Don’t confuse this with abstract, theoretical modeling — it’s a clear way to make ambiguity visible and tie definitions back to the business purposes they serve. Imagine a mind map where “Visitor” sits in the center and each department connects their meaning.

In fact, David Knickerbocker, author of Network Science with Python has shared excellent examples of how AI can help people interact with such graphs:

Why does this matter? Because graphs like this are super useful and serve more than pure documentation. They can actively power your AI applications. For instance, an AI assistant could query a graph and respond with the most relevant data based on a user’s role and access level:

3) Smart Schema Validation

Not all problems are definitional — some are just bad data. Test records, impossible timestamps, or ticket codes that don’t fit expected distribution patterns can pollute metrics in ways that mislead both teams.

AI can flag these anomalies automatically, keeping both definitions honest by ensuring they start with clean inputs. Think of it as an invisible team of data stewards, working around the clock in the background for you.

What AI Can't Solve

For all its upside, AI can’t do the hardest part: deciding what your metrics should actually mean.

AI can detect that marketing and operations measure "success" differently. It can even keep those definitions consistent across dashboards. But it can’t decide whether your strategy should prioritize lead volume or lead quality.

Should customer lifetime value include one-time buyers or only repeat customers? That’s not a technical question, but a business decision, rooted in whether you’re optimizing for growth or retention.

AI can also surface when the same executive uses one customer definition in a budget meeting and a different one in a board presentation. But it can’t resolve the incentives and organizational politics that drive that behavior.

What AI can do is force these conversations into the open. Instead of hiding behind technical fog or vendor black boxes, leadership has to confront the trade-offs explicitly.

That’s where the real progress happens.

The Practical Transformation

Here’s what this shift looks like in practice.

In the past, my simple visitor count example would take weeks to resolve. Teams dug through vendor reports, reverse-engineered BI dashboards, and sat in endless meetings just to untangle what the numbers actually meant. By the time we had clarity, the decision was late, and management moved on.

With AI-powered data management, that process could be flipped:

• Traditional approach: Discover conflicting numbers during quarterly reviews, spend weeks figuring out why they differ, have heated discussions about who's "right," eventually pick one definition by committee.

• AI-powered approach: Definitions are monitored continuously. Conflicts are flagged before they derail decision-making. Stakeholders see clear explanations of why numbers differ and are given explicit choices about the trade-offs. Definitions can stay “fluid,” applied differently depending on context, but always transparently.

My case happened over a decade ago, but I see the same definitional chaos in companies today.

The difference now is that AI has turned these hidden inconsistencies into visible business risks – and also given us new tools to solve them.

Thanks to the rise of Large Language Models, the technical backbone for leveraging AI in data management now exists. The question is whether your organization is ready for the transparency it brings.

Don't fight the fact that different teams want different metrics. Use AI to make those choices more visible – and more honest.

See you next Friday!
Tobias