What does a data scientist do?
By the Compass Careers team · Updated 6 November 2026
A data scientist uses data and statistical or machine-learning methods to answer questions that a simple report cannot, such as predicting what might happen or grouping similar customers. Much of the job is not the model itself. It is deciding whether a model is the right tool, getting the data into usable shape and checking whether the result can be trusted.
A typical week
- Framing the question: agreeing with a team what they want to know and how an answer would be used.
- Preparing data: cleaning records, handling gaps and deciding which fields matter. This often takes the largest share of the time.
- Trying methods: starting with a simple approach and moving to a more complex one only if it helps.
- Testing the result: checking the model on data it has not seen and looking for where it fails.
- Explaining it: telling non-technical colleagues what the output means and where it should not be used.
Compared with nearby roles
| Data scientist | Data analyst | Machine learning engineer | |
|---|---|---|---|
| Main question | What can we learn or predict from this data? | What do the numbers say about what happened? | How do we run a model reliably inside a product? |
| Typical output | Models, experiments, findings | Reports, dashboards, explanations | Production systems that serve a model |
| Heavier on | Statistics and experimentation | Questions and communication | Software engineering |
| Uncertainty | Often high: the answer may be "the data cannot tell us" | Moderate | Moderate: focus on reliability |
A decision to practise
A model predicts well on past data but you are unsure it will hold next month. Do you ship it, keep testing or explain the doubt and let the team decide? Try a free decision on Compass to notice what you reach for first. Nobody scores you.
Signs you might enjoy it
- You like finding out whether an idea is actually true.
- You are comfortable with maths and statistics, or keen to build that comfort.
- A "we are not sure yet" answer feels honest to you, not unsatisfying.
Signs it might frustrate you
- You want quick, visible results every day.
- Cleaning messy data feels like a waste of time.
- You would rather build the product than analyse it.
How to test it in a week
- Pick a small public dataset on a topic you care about, such as one from India's open government data portal.
- Write one question it might answer, then try the simplest method you know.
- Write down one way your answer could be wrong, and one check that would test it.
Common questions
Do I need a postgraduate degree?
Requirements vary by employer. Read the job descriptions you are aiming at, and look at what the team's work actually involves.
Where can I read more?
See what a data analyst does and data analyst vs data scientist.
Try the work before you choose it
Compass lets you try a realistic decision from each path you’re weighing, notice how it felt, and choose your next step with evidence. The preview on our home page is free.
Weighing two or three paths seriously? See our packages, each prepared and reviewed by a person.