The Data Science Workflow
The step by step that turns a vague question into a real answer.
Key Points
- Step 1: Ask a clear question. Not 'find insights' but 'does changing X increase Y?'
- Step 2: Get the data, from databases, CSVs, APIs, or scraping.
- Step 3: Clean it, fix missing values, standardise formats, remove duplicates.
- Step 4: Explore (EDA). Visualise, summarise, find weird stuff.
- Step 5: Model. Use stats or ML to answer your question.
- Step 6: Communicate. Share results so someone can actually use them.
What it is
It's just a repeatable process. You start with a question, then get data, clean it, explore it, build a model, and finally tell someone what you found. Real projects loop back. You learn something late that forces you to rethink earlier steps. That's fine. Normal.
Why it exists
Without a workflow, you just stare at numbers and try random things. That's not data science, that's guessing. A structured workflow saves time, reduces mistakes, and makes your work reproducible. Also, when you're stuck, you know what to do next.
Intuition
How it works
- Question definition
Turn a business problem into a data question with a clear success measure.
- Data acquisition
Find and pull the data you need, from internal databases, external APIs, or flat files.
- Data cleaning
Fix missing values, standardise formats, kill duplicates, handle outliers.
- Exploratory analysis (EDA)
Plot distributions, check correlations, generate guesses about patterns.
- Modelling
Build statistics or machine learning models to answer the question, then validate them.
- Communication
Present findings with visuals, plain language, and a clear 'so what?', not just code dumps.
When to use
- Every data project longer than an hour should follow some version of this.
- Use it to structure a new analysis from scratch.
- Use it to debug where things went wrong. Did you skip cleaning? Forget to explore?
Common mistakes
- Starting with modelling before looking at the data. Ouch.
- Treating the workflow as one way instead of looping back when you find issues.
- Leaving communication to the end. Thinking about your audience from the start changes everything.
Was this concept helpful?
Related concepts