Question first

The biggest mistake in data science isn't a coding error — it's starting without a clear question. "Let's explore the data and see what we find" feels productive but usually wanders into a swamp of charts that answer nothing anyone needed. Strong analysis starts with a specific, answerable question tied to a real decision: "Which customers are most likely to churn next month?" or "Did the new checkout flow increase completed orders?" A good question names what you're predicting or comparing, and hints at what you'd do with the answer.

What makes a question good

• Specific — "are sales okay?" is vague; "did weekday sales drop after the price change?" is testable.
• Answerable with data you can get — if the data doesn't exist, refine the question.
• Tied to a decision — know who acts on the answer and how. Analysis that changes no decision is trivia.

A repeatable loop

Once you have a question, real analysis follows a recognizable cycle:

1. Ask a specific question.
2. Get and clean the relevant data.
3. Explore and visualize it.
4. Model or analyze.
5. Communicate the answer clearly.

Expect to loop, not march

The steps look linear but rarely are. Early exploration reveals that your question was too broad, or the data can't support it, and you circle back and sharpen it. That's not failure — refining the question as you learn is the point. The one constant: keep the decision you're informing in view at every step, so you don't disappear down interesting-but-useless rabbit holes.

Analysis without a question is sightseeing. Anchor every project to a specific question and a real decision, and let the rest of the workflow serve that anchor.

Try this: Take a vague prompt like "look into our user data" and rewrite it as three specific, decision-linked questions. Notice how each one immediately implies which data you'd pull and which chart you'd make — that clarity is the whole value of asking well.