"Just Ignore It"

"Can't you just ignore it?"

I can't be the only governance professional who has heard this from a product owner or a developer. And every time I do, I have the same internal debate about sarcasm in the workplace.

"Oh sure, let's ignore it. That solves everything."

Too subtle and they might think you're serious. Too overt and suddenly you're having a very different kind of meeting — one that involves HR and an uncomfortable chair.

Become a governance professional, they said. It'll be fun, they said.

Here's the thing though. Every time someone says "just ignore it" — they're not making a data decision. They're making a governance decision. They just don't know it yet.

And that decision has a way of showing up later. In a boardroom where executives are making calls based on numbers nobody questioned. In an AI model's output that doesn't quite seem right — but nobody can explain why.

It will come up. You can count on it.

Benjamin Franklin once said "In this world nothing can be said to be certain, except death and taxes." He clearly never worked in data governance.

You have a column called EVENT_TYPE. Three values — documented in your data glossary, audited for quality. Clean.

Then the product starts sending a fourth.

Your automated quality check flags it. You ask the product team. The answer: "Oh, you can just ignore that."

Except you can't. That value is now in your production tables — tables you are responsible for. Nobody defined it. Nobody evaluated it for quality. Nobody confirmed it's real data and not a test record someone forgot to clean up.

An analyst could use it in a way that was never intended. Reports get built on it. Decisions get made from it.

Nothing is written down. (It never is.) The developer who added it knows what it is — and that knowledge lives entirely in their head.

So the questions start. How many values do you ignore before the spiral starts? What happens when the product adds a fifth? A sixth? At what point does "just ignore it" become the undocumented process?

It already is. You just can't see it.

So what happens when those undefined records make their way into an AI model?

AI doesn't stop when it hits something it doesn't recognize. It pattern-matches from context — the table name, the column name, surrounding values — and infers a definition. It states that inference with complete confidence. Nobody notices because the output looks reasonable. That's how hallucinations happen.

Now imagine those undefined event types are safety violations from a test environment that never got cleaned up. Driver IDs now have safety violations attached to them that never happened. You train a model on this data. The bias gets ingrained so deep you can't see it from the output. The model scores those drivers as high risk — confidently, consistently. HR acts on it. Drivers get fired.

All based on events that never happened. All because someone said "just ignore it."

That's a lawsuit waiting to happen.

Data without action is just overhead. If it's not meant to be used, don't store it. If you're storing it, know exactly what it is.

And never let AI define your data. The risk is too high.

Before you go:

Has someone ever told you to "just ignore it"? A bad value, a duplicate, a field nobody could explain? Email me at tj@guardianagentics.com

— T.J.