The Data Isn’t Wrong. You’re Just Missing the Rest of the Story.

Published: May 2025

By Amy Humke, Ph.D.
Founder, Critical Influence

The forecast is off.

The lift isn’t real.
The model was solid—until it wasn’t.

Not because the algorithm was flawed.
Not because the features were poorly selected.

It was because no one had told the model the context had changed—or that the context was never shared or understood.
This isn’t just a technical problem. It’s a visibility problem.
And it affects both doers and leaders.

The Context Problem No One Wants to Talk About

We talk a lot about model accuracy, drift detection, and hyperparameter tuning in data science. But one of the most dangerous problems is one we rarely mention out loud:
👉 Missing business context.

It happens when: - A marketing campaign launches mid-test and skews your outcome variable.
- A policy change alters behavior, but the model doesn’t know.
- A new incentive drives patterns that your model mistakes for fraud.
- A seasonality pattern is treated as a linear trend.

In each case, the model didn’t fail.
It was isolated.
The data scientist didn’t know.
The stakeholders didn’t think to mention it.
And the assumptions never got updated.

Why Knowing the Business Process Is Non-Negotiable

(A Theoretical Scenario)

At one hospital, complaints about radiology delays were mounting. The scheduling system estimated 45 minutes per MRI, based on thousands of past appointments. But something was off.

Appointments were consistently running 10–15 minutes long. It wasn’t catastrophic—but it caused patient backups, overworked techs, and delays across shifts.

Leadership brought in the data science team.
They reviewed the usual variables: patient age, scan type, machine model. Nothing explained the delays.

Until someone asked the people doing the work.

“It’s the piercings,” one MRI tech said. “If a patient shows up with a lot of jewelry and can’t remove it, we lose 15 minutes right there.”

The room went silent.

There was no field for “several piercings.” No flag for “requires extra prep.” But it mattered. A lot.

That insight didn’t come from a dashboard.
It came from the floor.

💡 If doers hadn’t spoken up, leaders wouldn’t have known what was broken.
💡 If leaders hadn’t created space to ask, that insight never would’ve surfaced.

Both sides mattered. And both sides had blind spots.

Predictive Power Without Process Awareness Is Just Guesswork

You can’t model what you don’t know exists—and this issue spans industries:

• Risk models might underestimate healthcare needs because they use historical spending as a proxy, unaware of past under-treatment in minority populations.
• Enrollment models might miss yield spikes after a last-minute tuition freeze the data team was never told about.
• Routing models might fail after a regulatory change affects drive-time policies, but the model still optimizes based on outdated rules.

These aren’t modeling failures.
They’re context failures—where the process, policy, or environment wasn’t in view.

Context Hides in Corners: It's Everyone’s Job to Surface It

For doers: - Don’t assume leadership knows what’s slowing you down. - Speak up about patterns that don’t show up in dashboards.

For leaders: - Don’t assume your analysts know about that new initiative. - Normalize sharing “non-data” details—those quirky edge cases that affect the numbers.

Context isn’t just a memo. It’s: - The workaround techs use when a system’s buggy
- The incentive that shifts behavior without touching the data
- The weather delay that throws off throughput

And often, no one thinks to mention it—until it’s already warped the results.

Staying Connected Isn’t About Location—It’s About Intention

Remote work isn’t the issue. Lack of communication is.
Being on-site doesn’t guarantee insight. Remote doesn’t mean disconnected.

What matters is intention. Are you creating space for insights to surface?

For doers, that means: - Asking “What’s changed?” regularly
- Staying curious beyond the dataset
- Probing for unseen impacts on business processes
- Building relationships that help edge cases bubble up

For leaders, that means: - Looping in data partners before changes happen
- Encouraging off-script conversations in meetings
- Asking what’s happening that isn’t written in the process
- Valuing real-world friction as a source of insight

That insight about piercings? It came from connection, not location.

Whether you're remote, hybrid, or on-site, modeling starts with understanding the system—by talking to the people inside it.

This Is Why Meetings Still Matter (Even When They're Imperfect)

Yes, meeting fatigue is real.
But so is context fatigue, when models break not from bad code—but from variables no one mentioned.

• For doers, tuning in to hear what’s changing (even informally) can save weeks of debugging.
• For leaders, sharing context early—even when it seems irrelevant—can prevent solving the wrong problem entirely.

Because the most critical variable might start as a comment someone almost didn’t say.

Better Models Begin With Shared Context

When insight flows both ways—up from the floor and down from the strategy—everyone wins:

✅ Feature selection becomes more relevant
✅ Feature engineering becomes more grounded
✅ Debugging becomes more efficient
✅ Cross-functional trust gets stronger

But that doesn’t happen by accident.
It takes deliberate communication, curiosity, and systems that reward transparency.

So What Do We Do?

For doers: - Get closer to the work, not just the data.
- Ask “What changed?” early and often.
- Follow the weird stuff—it usually matters.

For leaders: - Loop in analysts when things are still “just an idea.”
- Invite friction, not just status updates.
- Build a culture where surfacing edge cases is seen as value—not noise.

The Real Lesson

Even the best model can’t account for what no one shared.

Our job isn’t just to predict what happens next.
It’s to understand what shapes those outcomes—and that starts with conversation.

Because the model isn’t wrong.
You’re just missing the rest of the story.

📚 Want to See What Data Alone Can’t Show You?

Check out Anthro-Vision by Gillian Tett → https://amzn.to/4dujh6Y

Tett makes a compelling case for why organizations need more than dashboards—they need anthropologists, ethnographers, and listeners.

She shows how understanding behavior, culture, and context reveals the “missing pieces” that often explain why a model fails or a decision succeeds.

If you work at the intersection of data, people, and outcomes, this book belongs on your shelf.

Because sometimes, the most important variables don’t live in your dataset.

#datascience #analytics #businesscontext #featureengineering #communication #modelinterpretability #AIethics #leadership #remotework