Brandeis Online course developer and instructor Andrew Hua has been helping researchers, clinicians, faculty, and students use data to answer real-world questions for over 15 years. Hear his take on what’s important in data science today.
When people hear “data science,” they often think first about machine learning, predictive models, artificial intelligence (AI), or sophisticated analytical tools. Those are important parts of the field, but over the years I have come to believe that they are rarely where the most important work begins.
Effective data science starts with a much simpler question: What decision are we trying to make? That question may sound obvious. In practice, it is often one of the most difficult — and most valuable — questions a data professional can ask.
Start with the decision, not the algorithm
Throughout my career as a statistician, data scientist, consultant, and educator, I have collaborated with researchers, clinicians, faculty, students, and professionals from many different disciplines. A common pattern is that someone approaches a data scientist with a dataset and a method already in mind: “Can we run this model?” “Can we use machine learning?” or “Which algorithm should we choose?”
My instinct is usually to go one step backward.
What are we actually trying to understand? Who will use the result? What decision could change because of the analysis? What evidence would be useful enough to support that decision? Those questions often determine the analytical approach more effectively than beginning with a favorite technique.
It’s also a perspective I bring to my teaching in Brandeis Online’s MS in Applied Data Science and Decision Analytics, where students learn to connect analytical methods and technologies to the real-world decisions they are meant to support.
A technically correct answer can still be the wrong answer
One lesson I have learned through years of statistical consulting is that an analysis can be technically sophisticated and still provide limited value if the original question is poorly framed. A model might have strong predictive performance without addressing the problem an organization actually needs to solve. A statistically significant finding might be real but not practically meaningful. Likewise, an impressive visualization can communicate patterns clearly while still leaving the audience wondering, “So what should we do?”
This is why problem framing is not simply a preliminary step before the “real” analysis. Problem framing is part of the analysis.It helps define what success looks like, what data are needed, which assumptions matter, and how the eventual results should be interpreted.
Real-world data makes context essential
This became especially clear to me while working with complex healthcare and research data. Large datasets can create the impression that the analytical problem is primarily technical. In reality, the most consequential decisions often happen before the formal modeling begins.
For example, an analysis may require deciding who belongs in the study population, how a clinical or behavioral condition should be defined, which records should be included, how missing information should be handled, and whether variables actually represent what we think they represent. Those decisions can materially affect the final results.
In one healthcare research project, a substantial amount of work occurred before any formal modeling was performed. The team needed to define the study population, harmonize information from different data sources, clarify variable definitions, examine data quality, and make sure the analytical plan reflected the underlying research question.
None of that work was merely preparation for the analysis. It was part of the analysis. Experiences like this reinforced an important lesson for me: better data science often comes from better questions, clearer definitions, and stronger analytical judgment — not simply from more complicated models.
The best question often emerges through collaboration
Data scientists rarely develop the strongest analytical questions alone. A subject-matter expert understands the real-world context. A clinician or practitioner understands how a process actually works. An organizational leader understands which decisions need to be made. The statistician or data scientist contributes another perspective: what the available evidence can reasonably support. The strongest projects often emerge when these perspectives are brought together.
Over many years of consulting across healthcare, education, behavioral science, environmental research, linguistics, and other fields, I have seen the same pattern repeatedly. Productive projects usually involve an ongoing conversation between the substantive question and the analytical possibilities. Sometimes the most valuable contribution a data scientist can make is not immediately providing an answer. It helps the team refine the question until the answer actually matters.
Better decisions require more than prediction
Modern data science places considerable emphasis on prediction, and for good reasons. Predictive models can identify patterns that would be difficult or impossible to detect manually. But decision-making usually requires more than asking, “What will happen?” We may also need to ask: Why might it happen? How certain are we? Which assumptions affect the conclusion? Does the pattern hold across diverse groups?
And perhaps most importantly: What are the consequences of acting on this result, and what happens if the model is wrong? These questions bring together statistical reasoning, domain knowledge, data quality, interpretation, ethics, and communication. That is why I increasingly think of data science not simply as a collection of algorithms, but as a decision-making discipline supported by data.
AI makes analytical judgment more important, not less
The rapid development of artificial intelligence makes this distinction even more important. AI tools can now help generate code, summarize information, suggest analytical approaches, produce visualizations, and accelerate many parts of an analytical workflow. That creates significant opportunities for productivity.
It also makes it easier to produce an impressive-looking analysis before we have adequately considered whether the analysis is appropriate. Faster analysis does not automatically produce better decisions.
We still need to understand where the data came from, evaluate its quality, examine assumptions, validate results, document the workflow, communicate uncertainty, and explain how a conclusion was reached. In that sense, AI does not reduce the importance of analytical judgment. It may actually increase it. When tools make it easier to generate an answer, knowing which question deserves an answer becomes even more valuable.
From technical skills to professional judgment
I emphasize this distinction when teaching data science as well. Students naturally want to learn programming languages, statistical methods, visualization tools, machine-learning techniques, and emerging AI technologies. Those skills are important and necessary. But I also encourage them to develop the habit of asking what comes before and after the technical work.
Before the analysis: What problem are we solving? What decision are we supporting? What data would provide relevant evidence? After the analysis: What does the result mean? How certain are we? What are its limitations? How should we explain the findings to someone who was not involved in the analysis? Those habits are what transform technical capability into professional judgment.
Building data scientists who can help people decide
This perspective also reflects the broader purpose of Brandeis Online’s MS in Applied Data Science and Decision Analytics. Organizations certainly need professionals who can work with modern analytical methods and technologies. But they also need people who can connect those methods to real problems. That means developing professionals who can frame questions, evaluate evidence, recognize limitations, communicate results, and translate analysis into action.
The model matters. The code matters. The data matters. But ultimately, the purpose of data science is not simply to produce another model. It is to help people make better decisions.
This article was written by Andrew Hua, Applied Data Science and Decision Analytics course developer and instructor at Brandeis Online. Read more about Andrew on our Faculty and Advisory Council page.
