Behind the Hiring Desk · Data Analyst Careers 2026
What Companies Actually Want From a Data Analyst
Short answer: companies want a data analyst who can pull the right data with SQL, check that it is correct, turn it into a clear answer to a business question, and explain that answer to someone who does not care about the tools. Job descriptions list many tools, but the hiring process itself shows what is really being judged. This guide follows a typical analyst hiring process round by round and explains what each round is designed to find out, with an example of the kind of SQL task that decides many shortlists.
Processes differ between a bank, an e-commerce startup and an analytics services firm, so treat the rounds below as a common pattern, not a fixed rule. If you want to decode individual job postings first, read data analyst skills in job descriptions.
Round 0: the resume screen
What it checks: whether you are plausibly able to do the job, in the minute or two a recruiter or hiring manager spends on your resume.
Signals that help: projects described by the question they answered ("found which product categories drove repeat orders") rather than by tools used; a link to a portfolio or GitHub with readable work; SQL, Excel and one BI tool stated plainly. Signals that hurt: long tool lists with no evidence, generic objectives, and course certificates presented as experience. See tips for a strong data analyst resume for layout.
Round 1: the SQL and Excel test
What it checks: whether you can get correct numbers out of raw tables without help. This is often an online test or a live screen-share. Typical tasks: joins, GROUP BY with HAVING, handling NULLs, de-duplicating rows, and window functions for rankings and period-over-period change.
A representative task: "From the orders table, show each city's monthly revenue and the change from the previous month."
WITH monthly AS (
SELECT city, order_month, SUM(amount) AS revenue
FROM orders
GROUP BY city, order_month
)
SELECT city, order_month, revenue,
revenue - LAG(revenue) OVER (PARTITION BY city ORDER BY order_month) AS change
FROM monthly
ORDER BY city, order_month;
On a small sample of eight orders (run with SQLite through Python 3.12), the result is:
| city | order_month | revenue | change |
|---|---|---|---|
| Delhi | 2026-07 | 2000 | NULL |
| Delhi | 2026-08 | 1500 | -500 |
| Indore | 2026-07 | 400 | NULL |
| Indore | 2026-08 | 1100 | 700 |
| Pune | 2026-07 | 900 | NULL |
| Pune | 2026-08 | 1300 | 400 |
What the reviewer looks for is not only that the query runs. They check whether you aggregated before applying LAG (otherwise you compare individual orders, not months), whether you partitioned by city (otherwise Indore's July is compared with Delhi's August), and whether you can explain why the first month is NULL and what you would show a manager instead. Those three points are the difference between "knows SQL syntax" and "can be trusted with data". The SQL commands every analyst needs covers the building blocks.
Excel questions in the same round usually involve lookups (XLOOKUP or INDEX/MATCH), pivot tables, and cleaning a messy column of dates or names.
Round 2: the case study or take-home task
What it checks: whether you can go from a vague business question to a defensible answer. A common format: a CSV of orders, users or support tickets, and a prompt such as "Repeat purchases fell last quarter. Find out why and recommend what to do."
Strong submissions tend to share a shape:
- Restate the question and say how you will measure it (for example, "repeat rate = customers with two or more orders in 90 days").
- Check the data first: duplicates, missing values, date ranges, test orders. Say what you removed and why.
- Break the metric down by the dimensions that could explain it: city, acquisition channel, first-order category, delivery time.
- State the finding with numbers from the data, and its limits ("this is correlation; an A/B test would confirm it").
- Recommend one or two actions a manager could take next week.
Weak submissions usually have the opposite pattern: many charts, no single conclusion, and no mention of data quality.
Round 3: the business or hiring-manager conversation
What it checks: judgement and communication. Expect questions such as:
- "Our app's daily active users dropped 8% yesterday. What would you check first?" (The number is part of the question, not a fact.) Good answers start with data issues such as tracking changes or a missing day of data, then move to segments such as platform, app version and city, then to external causes like a holiday.
- "Which metric would you use to measure success for this new feature, and what could go wrong with it?"
- "Tell me about a time your analysis was wrong." They want to hear how you found the mistake and what check you added afterwards.
Some companies, especially consulting and e-commerce firms, add a guesstimate here, such as estimating the number of food delivery orders in a city per day. The point is a structured approach, not the final number.
Round 4: culture, HR and practical details
What it checks: how you work with others and whether the role suits you: working with stakeholders who change requirements, meeting deadlines on recurring reports, comfort with documentation. Be honest about what you have and have not done; analysts are trusted for accuracy, and overclaiming in an interview works against exactly that.
What sits behind all the rounds
| What companies want | Where it is tested | How to show it |
|---|---|---|
| Correct numbers | SQL test, take-home | Data checks written into every project |
| Business framing | Case study, manager round | Projects that start from a question, not a dataset |
| Clear communication | Presentation, manager round | A one-page summary with each portfolio project |
| Reliability | References, HR round, probation | Reproducible work: saved queries, versioned files |
| Willingness to learn tools | All rounds | Depth in one BI tool, not a list of six |
Python with pandas is increasingly expected for analyst roles beyond pure reporting, and basic statistics (averages vs medians, distributions, significance) appears in many interviews. Neither replaces SQL and clear thinking.
Preparing in the right order
Practise SQL until window functions and joins feel routine. Build two or three portfolio projects that each answer a business question and include a short written summary; portfolio projects for data analyst roles has a template. Do one timed take-home on a public dataset to see how long a clean submission really takes you. Before interviews, read common data analyst interview mistakes and compare yourself honestly against what a job-ready analyst can do.