The Skill That Separates Senior Data Scientists From Everyone Else

Sebastian Wernicke on communication, stakeholder resistance, and why data alone rarely changes minds.

I sat down with Sebastian Wernicke on the ODSC AiX Podcast to talk about the skills that actually move a data scientist’s career forward.

Most of them are not technical.

Wernicke leads data science teams out of Munich, came up through bioinformatics, spoke at ODSC AI East, and recently published Data Inspired. His core argument is simple: the modeling is usually the part data scientists are already good at. The work that gets them to the next level is the work they often avoid.

That work is communication. It is stakeholder management. It is reading incentives, politics, fear, and resistance inside an organization. It is learning how to make data useful to people who do not think like data scientists.

And in the age of AI, that skill set is becoming more important, not less.

The Problem Is Not Always a Data Deficit

Data scientists often assume that better data leads to better decisions.

Bring the right dataset to the right people at the right time, and the organization will make the rational choice. That is the story many technical teams want to believe.

Wernicke pushed back on that assumption.

“It turns out that psychological research for decades has shown that data does not change our minds,” he said. “In the worst case, it actually does the opposite.”

The stronger someone’s belief, the more likely they are to discount evidence that contradicts it. That is exactly when data should matter most, but it is also when people are most likely to resist it.

Wernicke calls this the “data-resistant mind,” which is both the subject of his ODSC talk and a chapter in Data Inspired. Inside an organization, that resistance becomes even more complicated. Data does not arrive in a neutral environment. It arrives inside hierarchy, politics, budgets, job security, competing incentives, and personal reputation.

Kodak, Blockbuster, and BlackBerry all had clear signals that their markets were changing. The issue was not a complete lack of data. It was what people were able, or willing, to do with the data they already had.

As Wernicke put it: “It’s not a lack of data, it’s how we process that data.”

That is where the data scientist’s real job begins.

A clean model is not enough if it lands in front of people who are wired, incentivized, or pressured to reject its implications.

Communication Is Not a Soft Skill

Wernicke runs communication training sessions for the data scientists he leads. At first, many resist it. They expect training on neural networks, the latest architecture, or another technical tool.

His answer is direct.

“You already know neural networks. You know AI. XGBoost. You need to learn how to communicate to the stakeholders and how to listen to them because that’s going to be your crucial skill that gets you to the next level.”

That point matters because “communication” is often treated as a vague professional virtue. Everyone agrees it is important, but few people define what it actually means.

For Wernicke, communication starts with translating the data into the language of the person across the table.

Take a model output that says there is a 66 percent chance of success. To a data scientist, that number may feel precise. To a decision-maker whose budget, product launch, or career is on the line, it may feel like a coin flip with no obvious next step.

“A decision-maker doesn’t know what to do with that,” Wernicke said.

His fix is not to hide the uncertainty. It is to make the uncertainty usable.

Instead of presenting one number, describe the scenarios. Show the paths where the launch fails. Show the path where it works. Explain what signals would tell the business which path it is on. Then the same data becomes something a stakeholder can actually process.

“You’re presenting the same data,” he said, “but you’re presenting it in the language of the stakeholder that you’re speaking to.”

That is not dumbing the work down.

It is making the work actionable.

The Objection Is Not Always the Objection

Communication also means understanding why a stakeholder is resisting you.

The stated objection is often not the real one.

Wernicke shared a story about a client project where one of his colleagues solved a classification problem cleanly and effectively. The issue was that the solution did not use AI.

That should have been fine. The client’s problem was solved.

But the project manager demanded that Wernicke fire the colleague.

Why? Because the client had secured a budget specifically earmarked for AI. A successful non-AI solution created an internal problem. It exposed the project manager to questions about whether the budget had been used properly.

The model was not the issue. The incentive structure was.

The work succeeded, but the relationship nearly broke for reasons that had nothing to do with classification accuracy.

This happens more often than data teams like to admit. A stakeholder who blocks progress may be protecting a budget. A manager may have already lost support from above. A team may be afraid that the model will make their work look less valuable. A department may resist because the recommendation shifts power somewhere else.

Wernicke’s advice is to map the stakeholders before presenting.

Who are they?
What do they need?
What are they afraid of?
What incentives are shaping their reaction?
Where might hidden resistance appear?

Some of that resistance is irrational. Some of it is entirely legitimate. Either way, ignoring it does not make it disappear.

Optimize or Transform?

The same communication problem shows up when data scientists propose what the business should do next.

Wernicke separates two uses of data: optimizing the existing business and transforming what the business does.

He ran into this distinction on a project for a German streaming service. The client wanted a traditional audience segmentation model. The goal was to infer each viewer’s age, gender, and demographics from their viewing habits.

Wernicke pushed back.

The company already knew what each viewer actually watched. That behavioral data was more useful for selling ad slots than a demographic label designed for an older media world.

“You’re asking us to optimize this very old metric,” he told them.

The better move was not to improve the demographic proxy. It was to stop relying on the proxy and sell against real viewing behavior, closer to performance marketing.

He lost the argument.

The client was too attached to the old segmentation model to move.

That story captures one of the harder truths of data science: the better answer does not win automatically. A stronger model, cleaner metric, or more modern approach is not enough if the organization is not ready to hear it.

This is why senior data scientists are not just better at algorithms. They are better at timing, framing, and organizational judgment.

What AI Changes About the Data Science Role

Wernicke is clear-eyed about how quickly AI is changing the technical side of data science.

He described an optimization problem that took him three weeks to code last year. This year, using a current AI model, it took two hours — and the model produced a better result.

He described the feeling as vertigo: fascinating and concerning at the same time.

His view of the future comes from an older split in programming work. Some programming jobs were based on receiving precise instructions and translating them into code. That work became easier to commoditize. Other roles required taking a messy business problem, translating it into a technical problem, solving it, and translating the answer back into something the business could use.

Those roles held more value.

Wernicke expects data science to split in a similar way.

“If you’re viewing data science in a very narrow sense, as in, ‘I want somebody to give me my clean data set and then run XGBoost on that,’ that is going to go away.”

The narrow version of the job is under pressure.

The broader version is not.

The data scientist who can understand a business problem, ask better questions, choose the right method, explain uncertainty, read stakeholder resistance, and guide a decision still has a valuable role. AI may change the tools, but it does not remove the need for judgment.

If anything, it raises the bar.

Why AI Makes Trust More Complicated

There is another wrinkle.

Wernicke pointed out that people often distrust traditional machine learning systems. One visible mistake can be enough for a stakeholder to dismiss the model, even if it outperforms humans on average.

AI creates a different problem.

Because generative AI is fluent, people often trust it too quickly. They tell it about their health, their colleagues, their strategy, and their private concerns. They cite it as an authority in arguments. It speaks in a way that feels confident, helpful, and human enough to earn trust it has not always earned.

“From an algorithmic standpoint, something that we should be very distrusting of,” Wernicke said, “but at the same time it pushes exactly the right buttons to create trust.”

His conclusion is not to avoid AI. He advocates using it.

But he argues that teams need to understand both sides: the technical limits of AI systems and the interaction dynamics they create. A tool that sounds confident can shape decisions even when it is wrong. That makes communication, skepticism, and judgment even more important.

The Advice for Data Scientists

When I asked Wernicke what data scientists should take away, he brought the conversation back to the same point.

“As data scientists you are very much in a comfort zone of data,” he said. “If you are looking to make that next step in your career, it’s just important that you learn about the other aspects of a dynamic in an organization. So about things like power, things like politics, things like communication, but communication’s usually the entry drug.”

That phrase stuck with me: communication is the entry drug.

It is the first step into the broader set of skills that separates a strong individual contributor from a senior data scientist, team lead, or trusted advisor.

Technical depth still matters. It always will. But technical depth alone is no longer enough, especially as AI makes parts of the technical workflow faster and more accessible.

The data scientists who keep growing will be the ones who can do the full job: understand the data, understand the model, understand the business, and understand the people making the decision.

As Wernicke put it: “If you can bring that whole package, then I think you have great times ahead of you, no matter what’s happening on AI and everything else in the world.”

That is the real lesson.

The future of data science will not belong only to the people who know the newest model.

It will belong to the people who can help organizations think better.