What analysts do

The job before the tools: turning a vague business worry into a question, deciding how it will be measured and choosing the kind of analysis that answers it.

What a data analyst does

Also called data analyst, BI analyst, business analyst

What: Answers business questions with data: finds and cleans the data, analyses it, and explains what it means and what to do next, usually through reports, dashboards and short write-ups.
Why: Knowing the job keeps you from over-investing in tools. Employers pay analysts for decisions they improve, not for queries or charts on their own.
Example: A delivery company asks why late deliveries rose in March. The analyst pulls delivery records, finds that one warehouse changed shifts, and shows the effect in one chart.

Watch out

Thinking the job is making dashboards, and never asking who will use them or what decision they support.

A good business question

Also called business question, problem statement, analysis question

What: A specific, answerable question tied to a decision: who is asking, what they will do with the answer, which metric, which period and which comparison.
Why: Most wasted analysis starts from a vague request. A sharp question decides what data you need and when you are finished.
Example: "How are sales?" becomes "Did the June discount raise repeat orders among customers in their first 90 days, compared with May?"

Watch out

Starting to query before agreeing the question, then answering a different one from the one the stakeholder had in mind.

Metrics and KPIs

Also called KPI, key performance indicator, north star metric

What: A metric is a precisely defined number, such as weekly active users. A KPI is one of the few metrics a team has chosen to judge its performance by.
Why: If two teams define "active user" differently, their dashboards disagree and nobody trusts either. Written definitions are the foundation of every report.
Example: Active customer: placed at least one paid, non-refunded order in the last 30 days, counted in the business time zone.

Watch out

Leaving edge cases undefined (refunds, test accounts, time zones) so the same metric changes depending on who calculates it.

Types of analysis

Also called descriptive, diagnostic, predictive, prescriptive, descriptive analysis, diagnostic analysis, predictive analysis, prescriptive analysis

What: Descriptive says what happened, diagnostic asks why, predictive estimates what will happen, and prescriptive suggests what to do.
Why: Naming the type sets expectations: most analyst work is descriptive and diagnostic, and promising a prediction from a quick report sets you up to fail.
Example: Revenue fell 8 percent (descriptive) because repeat buyers dropped after a price change (diagnostic).

Watch out

Presenting a trend line as a forecast without saying what assumptions it rests on.

The analysis workflow

Also called analysis workflow, ask, prepare, process, analyse, share, act

What: The loop most analyses follow: ask the question, prepare and clean the data, analyse it, share the result and help people act, then check whether the action worked.
Why: Following the steps stops you from skipping cleaning or sharing, which is where most analyses go wrong or go unused.

Watch out

Spending all the time on analysis and ten minutes on the write-up that people actually read.

Tidy data

Also called long format, one row per observation

What: A table where each row is one observation, each column is one variable and each cell holds one value.
Why: Spreadsheets, SQL, BI tools and pandas all work best on tidy tables. Reshaping messy layouts into this form is a large part of cleaning.
Example: Instead of one column per month (Jan, Feb, Mar), use two columns: month and sales, with one row per product per month.

Watch out

Keeping totals, notes and merged header cells inside the data table, which breaks every tool downstream.

Personal data and privacy

Also called PII, anonymisation, anonymised, GDPR, personal data

What: Information that identifies a person, such as names, phone numbers, emails or national ID numbers, and the rules for collecting, storing and sharing it.
Why: Analysts see a lot of personal data. Mishandling it can break the law, harm people and end a career.

Watch out

Putting real customer or employer data in a public portfolio, a public dashboard or an AI chat, even "just to test".

Go deeper