First Principles Thinking for Data Scientists

wrote to the Smithsonian Institution asking for everything they had on human flight. He and his brother Orville devoured every paper, theory, and calculation from the leading aeronautical researchers of their time. They studied Otto Lilienthal’s glider designs, Octave Chanute’s engineering principles, and Samuel Langley’s tables on lift and drag.

Then they did something radical: they questioned all of it.

When their gliders didn’t perform as the published data predicted, the Wright brothers didn’t assume they were at fault. They built their own wind tunnel and tested over 200 wing designs themselves. What they found then stunned them. Lilienthal’s lift coefficients—the numbers the entire field had relied on—were wrong, and not just a little bit. Totally, completely wrong.

The Wright brothers didn’t question this out of arrogance; they were employing first principles thinking. And it’s what made the difference between their first flight in 1903 and Langley—who had more funding, prestige, and resources—crashing his aircraft into the Potomac just nine days earlier.

Today’s data scientists face a similar challenge. We’re surrounded by playbooks, frameworks, and best practices—A/B testing guides, metrics taxonomies, model selection checklists. These tools are invaluable. But just like Langley’s tables, they can lead us astray when applied blindly.

The most strategic data scientists I know don’t choose between frameworks and first principles. They use both. Frameworks give speed. First principles provide clarity. Together, they separate task executors from strategic partners.

What Is First Principles Thinking?

First principles thinking means breaking a problem down to its fundamental truths and rebuilding from the ground up. Aristotle described it as finding “the first basis from which a thing is known.”

In practice, it means asking:

• What do we know to be absolutely true?
• What are we assuming?
• Which of those assumptions can we challenge?

This doesn’t mean rejecting existing knowledge: the Wright brothers studied all the research available; they just didn’t treat it as gospel.

For data scientists, the same applies. Frameworks are maps—proven shortcuts through familiar terrain. First principles are the compass, keeping us oriented when the map gets a bit blurry.

Why Data Scientists Need This Now

Frameworks are everywhere in data science for good reason. They help us run experiments, define metrics, and build models quickly. But they can also create false confidence.

I’ve seen teams run flawless A/B tests that answered the wrong question. I’ve seen standard metrics produce dashboards that looked impressive but measured nothing meaningful. These failures didn’t happen because the frameworks were flawed. They happened because no one paused to ask the first principles questions: What decision are we actually trying to inform? What value are we really trying to measure? Do we even need this level of complexity?

This matters more than ever because AI is automating the execution side of data science. Gen AI can query data, generate visualizations, and apply frameworks brilliantly. But it cannot decide whether you’re asking the right question.

First principles thinking is your differentiator. It’s the skill that keeps frameworks grounded in reality—and it’s becoming the most defensible capability of a strategic data scientist.

Where First Principles Changes Everything

Here are three examples where the data science playbook is thoroughly well written, but strictly following it mindlessly would produce a poor outcome.

Beyond Textbook A/B Tests

The framework says: define your hypothesis, randomize users, measure your primary metric, check significance. That works perfectly—if you’re asking the right question.

But the first principles question comes earlier: What decision are we trying to inform? What uncertainty are we trying to resolve?

I once advised a team who was testing a new recommendation algorithm. Framework thinking said: randomize users, measure click-through rate, run for two weeks. Ship the winner.

But a first principles pause revealed something different. We weren’t uncertain about clicks—early signals suggested they would go up. We were uncertain whether those clicks would drive true engagement or just noise.

So we changed what we measured. Instead of clicks, we focused on return visits, session depth, and long-term engagement. The result? The new algorithm increased clicks by 12% but decreased return visits by 8%. The standard framework would have said “ship it.” The first principles approach said “not yet.”

We decided that the new algorithm was too “click-baity”. Frameworks gave us the methodology. First principles gave us the right question.

What Are Our Metrics Really Measuring?

Metrics frameworks—North Star, OKRs, HEART—are powerful because they give structure. But they can also create the illusion that we’ve measured what matters.

First principles thinking asks: What’s the fundamental behavior or value we care about? Does this metric actually capture it?

Consider engagement. Many frameworks suggest DAU, session length, or actions per session. Reasonable proxies—but are they right?

• For a meditation app, longer sessions may look “better,” but the fundamental goal is a sustainable practice. That might mean shorter sessions over time.
• For an analytics tool, more queries per user could signal deeper usage, or it could mean users are struggling to find answers. The real value is faster, more targeted insights.

I once started a new job and inherited a dashboard that proudly reported weekly active users as the primary success metric. But when I dug in, I realized most “active” users were just logging in, looking around, and leaving without completing a single task. From a first principles lens, I decided that the real value was tasks completed. And once I shifted the metric, we discovered (as expected) that usage was far lower under the new definition, but the new framing provided us clarity on what to focus on in order to drive meaningful adoption.

The framework gives you a menu of metrics. First principles thinking tells you whether any of them actually reflect your product’s value. Sometimes the standard metric is perfect, but sometimes it’s dangerously misleading.

When First Principles Saved a Launch

One of the clearest examples I’ve experienced came early in my career when my team was tasked with building a “user quality score” to help sales prioritize leads.

The framework approach was obvious: supervised learning, predict conversion probability, rank by score. We had the data, the features, the methodology.

Two weeks in, as we were struggling to eke out incrementally higher performance scores, someone asked: What decision will sales actually make with this score?

We asked sales. The answer wasn’t “give me exact probabilities.” It was: Should I take the time and personalization to call this lead, or just send a quick standardized email?

That shifted everything. We didn’t need a complex model spanning the full probability range. We needed a simple, interpretable classifier optimized around one threshold.

By reframing, we moved from an ensemble model to logistic regression, dropped half our features, shipped three weeks faster—and delivered something sales actually used.

By going back to first principles, we clarified the real problem and then went back to the standard frameworks to build a solution.

The Compass and the Map

Here’s the lesson: strategic data scientists don’t choose between frameworks and first principles. They combine them.

• Frameworks are the map—they let you move fast and leverage accumulated knowledge.
• First principles are the compass—they keep you aligned when the map isn’t clearly showing your path.

The Wright brothers didn’t reject the research of their time. They built on it, but they also knew when to go back to basics.

That’s the mindset shift that separates strategic data scientists from tactical ones. It’s not about knowing more methods or working harder. It’s about knowing when to follow the map and when to check your compass.

With the guidance of AI, frameworks will keep getting easier to apply. But the compass—that’s yours to build. And it’s what will keep you relevant, strategic, and indispensable in the years ahead.

This is one of the core themes I explore in my new book, The Strategic Data Scientist: Level Up and Thrive in the Age of AI (Amazon affiliate link). It’s about pairing foundational thinking with proven frameworks to create impact, influence roadmaps, and position yourself as a strategic partner—not just a technical executor.

If you’re wondering how your job will change as AI becomes more capable, or simply looking to have more impact as a data scientist and make progress towards a promotion, please check out the book on Amazon!