---
title: Data Analyst Interview Questions: 30 Answers and a Case
url: https://www.masaischool.com/resources/data-analyst-interview-questions
description: Data analyst interview questions for India: the rounds, 30 model answers on SQL, Excel, statistics, Power BI and cases, two worked examples and a 2-week plan.
updated: 2026-11-17
publisher: Masai (https://www.masaischool.com)
---

# Data Analyst Interview Questions in India: Rounds, 30 Model Answers, a Worked Case and a 2-Week Plan

> **Short answer**
>
> A data analyst interview in India usually combines a screening call, an SQL test, an Excel or Power BI task, a statistics round, a business case and a behavioural round, though no employer publishes a standard format. These 30 model answers cover SQL, Excel, statistics, visualisation, cases and behaviour, with two worked examples computed in Python and a two-week plan.

This page covers what a data analyst interview looks like in India, 30 questions with model answers, two worked examples and a two-week plan. The numbers in the worked examples are invented for practice, and both Python blocks ran on Python 3.9 using only the standard library. Tool facts come from Microsoft's documentation and the other sources listed, read on 11 November 2026.

Interest in SQL, Excel and Power BI in Google Trends India peaks in February, so start in January. For depth on the technical rounds, use the [SQL interview questions](https://www.masaischool.com/resources/sql-interview-questions) and the [Python interview questions](https://www.masaischool.com/resources/python-interview-questions) pages.

## What a data analyst interview round looks like in India

No employer publishes a standard format. Amazon's careers page says its process differs from role to role and lists an online application, assessments, phone screening and an interview loop. The table is our synthesis of what freshers and early-career candidates prepare for, not survey data. Masai's [data analyst interview questions for freshers](https://www.masaischool.com/blog/data-analyst-interview-questions-freshers-must-prepare-in-2026/) describes a similar four-round funnel.

| Round | What it checks | How to prepare |
|---|---|---|
| Screening call | Tools, projects, notice period, salary range, role fit | A one-minute pitch, two projects with numbers, [salary research](https://www.masaischool.com/resources/data-analyst-salary-in-india) |
| SQL test | Joins, aggregation, window functions, NULLs | Q1 to Q4 here, then the 40 SQL questions |
| Excel or Power BI task | Cleaning, pivots, lookups, a small dashboard | Q5 to Q14 |
| Statistics round | Distributions, p-values, intervals, A/B tests | Q15 to Q20 |
| Case study | A metric drop, funnel, retention or experiment, structured aloud | Q21 to Q26 and the worked case |
| Stakeholder round | Communication, pushback, prioritisation | Q27 to Q30 |

## SQL and Excel questions (Q1 to Q9)

- **Q1. How do you find the top 3 products by revenue in each month?** Aggregate revenue by month and product, then rank within each month with DENSE_RANK or ROW_NUMBER and keep rank 3 or less in an outer query, because window functions cannot be filtered in WHERE. Use DENSE_RANK if tied products should all count.
- **Q2. How do you find customers who ordered in January but not in February?** Use an anti-join: take January's customers and keep those with no February order, with NOT EXISTS or LEFT JOIN ... IS NULL. Avoid NOT IN when the subquery column can hold NULL, because one NULL empties the result.
- **Q3. How do you handle duplicates and NULLs before analysis?** Define a duplicate first (same order id, or same customer, date and amount), count them with GROUP BY ... HAVING COUNT(*) > 1, then keep one row with ROW_NUMBER. Ask whether a NULL means zero, unknown or not applicable before using COALESCE, because AVG skips NULLs and zeros change the answer.
- **Q4. When do you use SQL, Excel or Python?** SQL to pull and aggregate data where it lives. Excel for small files, quick pivots and sharing: a worksheet holds at most 1,048,576 rows. Python for repeatable cleaning, statistics and automation. Name the one you would pick for the task in front of you. See [free data analytics tools](https://www.masaischool.com/resources/tools/free-data-analytics-tools).
- **Q5. VLOOKUP, INDEX and MATCH, or XLOOKUP?**VLOOKUP needs the lookup value in the first column and defaults to approximate match unless you pass FALSE. INDEX with MATCH can look left, and referencing the return column directly means an inserted column does not break it. XLOOKUP is simpler, but Microsoft says it is not available in Excel 2016 or 2019.
- **Q6. What is a pivot table?** A summary of rows by categories, such as sales by region and month, without formulas. Refresh it when the source changes. Keep the source as a clean table with one header row and no blank rows or merged cells.
- **Q7. What do COUNTIFS and SUMIFS do?**COUNTIFS counts rows that meet several conditions. SUMIFS sums one column under conditions, and its sum range comes first: =SUMIFS(D2:D500, B2:B500, "North", C2:C500, ">1000").
- **Q8. How do you clean messy data in Excel?** Use Power Query, which Microsoft says saves transformation steps as a repeatable query you can refresh: trim, split, change types, remove duplicates, unpivot. TRIM and CLEAN suit one-off fixes. Manual edits do not rerun on next month's file.
- **Q9. How do you build a simple dashboard in a spreadsheet?** Keep three layers: a clean data sheet, a calculation sheet of pivots or formulas, and a presentation sheet with a few KPIs, charts and slicers. Label units and the date range. Our [Google Sheets and Looker Studio guide](https://www.masaischool.com/resources/build/dashboard-google-sheets-looker-studio) walks through one.

## Power BI and visualisation questions (Q10 to Q14)

- **Q10. What is the difference between a measure and a calculated column?** A measure is a DAX formula evaluated at query time in the current filter context, and Microsoft says its values are never stored in the model. A calculated column computes a result for every row and recalculates on refresh. Use measures for totals and ratios, columns for row-level labels.
- **Q11. What is a star schema, and why does Power BI prefer it?** Fact tables hold events such as orders. Dimension tables describe products, dates and customers. Microsoft says dimensions enable filtering and grouping while facts enable summarisation, which matches how report visuals query a model. Relate them one-to-many.
- **Q12. Import or DirectQuery?** Import loads data into Power BI's in-memory model: fast and fully interactive, but it needs refreshing. DirectQuery leaves data in the source and queries it for each visual, which suits constantly changing or uncopyable data at the cost of speed. Microsoft's guidance is to use Import by default.
- **Q13. How do you choose a chart?** Match it to the question: lines for change over time, bars to compare categories, scatter for relationships, tables for exact values. Avoid 3D effects and truncated bar axes. Sort bars, label units and put the key number in the title.
- **Q14. A dashboard is slow or shows odd numbers: what do you check?** For speed, look at visual count, heavy measures, model shape and the source; Microsoft points to Performance Analyzer to isolate slow visuals. For odd numbers, check filters, slicers, relationship direction, duplicates and grain before touching the DAX.

## Statistics, p-values and A/B tests (Q15 to Q20)

- **Q15. When do you use the median instead of the mean?** When data is skewed or has outliers, such as order values or delivery times, because a few large values pull the mean. Report the median with the spread, and say which you used.
- **Q16. What is a p-value?** The American Statistical Association says p-values can indicate how incompatible the data are with a specified model, and do not measure the probability that the hypothesis is true. In plain words: if there were no real difference, how surprising is data this extreme?
- **Q17. What does a 95% confidence interval mean?** NIST says that repeating the sampling gives intervals that bracket the true value in about 95% of cases. It is not a 95% chance that this one interval holds the truth. A wide interval means low precision.
- **Q18. What are Type I and Type II errors, and power?** Type I is a false positive, controlled by the significance level (often 5%). Type II is a missed real effect. Power is the chance of detecting a real effect, commonly set at 80%, and small samples lower it.
- **Q19. How do you run a trustworthy A/B test?** Fix the metric, the smallest effect worth detecting and the sample size first, randomise users, then test once. Check for sample ratio mismatch: Microsoft's research found about 6% of its tests had one and says to distrust such results until diagnosed. Evan Miller shows peeking can push false positives to 26.1% in his simulation.
- **Q20. Control 520 of 10,000 converted and the variant 585 of 10,000. What do you say?** The script computes p = 0.044, so significant at 5%, with a lift from 5.20% to 5.85%. But the interval for the difference is 0.02 to 1.28 percentage points and power is only 52%. Report the range, not just "significant", and consider a larger rerun.
from math import sqrt from statistics import NormalDist nd = NormalDist() n_c, x_c = 10000, 520 # control: users, conversions n_v, x_v = 10000, 585 # variant p_c, p_v = x_c / n_c, x_v / n_v diff = p_v - p_c # Two-proportion z-test: the standard error assumes no real difference p_pool = (x_c + x_v) / (n_c + n_v) se_null = sqrt(p_pool * (1 - p_pool) * (1 / n_c + 1 / n_v)) z = diff / se_null p_value = 2 * (1 - nd.cdf(abs(z))) # 95% confidence interval for the difference se = sqrt(p_c * (1 - p_c) / n_c + p_v * (1 - p_v) / n_v) z_a = nd.inv_cdf(0.975) low, high = diff - z_a * se, diff + z_a * se # Power of this test for the observed lift, and the sample size per arm for 80% power power = nd.cdf(diff / se - z_a) z_b = nd.inv_cdf(0.80) n_needed = (z_a + z_b) ** 2 * (p_c * (1 - p_c) + p_v * (1 - p_v)) / diff ** 2 print(f"conversion {p_c:.2%} vs {p_v:.2%}, lift {diff / p_c:+.1%} relative") print(f"z = {z:.2f}, p-value = {p_value:.3f}") print(f"95% CI for the difference: {low * 100:.2f} to {high * 100:.2f} percentage points") print(f"power at {n_c} per arm: {power:.0%}; users per arm for 80% power: {n_needed:.0f}")

Output:conversion 5.20% vs 5.85%, lift +12.5% relative z = 2.01, p-value = 0.044 95% CI for the difference: 0.02 to 1.28 percentage points power at 10000 per arm: 52%; users per arm for 80% power: 19390

## Business cases: metric drop and funnel (Q21 to Q26)

- **Q21. Orders fell 12% last week. How do you investigate?** First confirm the drop is real: tracking, pipeline, partial week. Then size it, split it by channel, device, region and new versus returning users, walk the funnel, and check releases, pricing, outages and holidays. In the numbers below, sessions were flat and the Android app explained 95.7% of the drop.
- **Q22. What does funnel analysis tell you?** Step-to-step conversion, compared across periods and segments. The biggest absolute drop is not always what changed. Here three steps held at 70.0%, 30.0% and 23.8%, and only checkout to order fell, from 60.0% to 40.0%. Next checks: payment errors, app version, gateway success rate.
- **Q23. How do you define success for a new feature?** One primary metric tied to the goal, such as completed orders per active user, a few guardrails that must not fall (refunds, load time, support tickets) and a time window. Avoid vanity metrics such as page views alone.
- **Q24. What is cohort analysis?** Group users by when they joined, then track each group over time, for example week 1 and week 4 retention. It separates older users leaving from newer users being worse. Plot cohorts as rows and weeks as columns.
- **Q25. Revenue is up but profit is down: why?** Split revenue into price, volume and mix, then look at costs: discounts, returns, shipping and acquisition. A shift towards low-margin products or heavier discounting is one possible cause. Ask how the company defines margin first.
- **Q26. How do you estimate a number with no data, such as daily orders in a new city?** State assumptions step by step (population, share who shop online, orders per person per month), compute, then sanity-check against a comparable city. Interviewers grade the structure more than the final figure.
# Invented numbers for practice, not real company data. # segment: (sessions before, orders before, sessions after, orders after) weeks = { "Android app": (40000, 1200, 41000, 820), "iOS app": (25000, 900, 25500, 918), "Mobile web": (30000, 600, 29000, 580), "Desktop": (15000, 450, 14500, 435), } sessions_a = sum(v[0] for v in weeks.values()) orders_a = sum(v[1] for v in weeks.values()) sessions_b = sum(v[2] for v in weeks.values()) orders_b = sum(v[3] for v in weeks.values()) print(f"Orders {orders_a} -> {orders_b} ({(orders_b - orders_a) / orders_a:+.1%})") print(f"Sessions {sessions_a} -> {sessions_b}, conversion {orders_a / sessions_a:.2%} -> {orders_b / sessions_b:.2%}") # Step 1: which segment explains the drop? drop = orders_b - orders_a for name, (sa, oa, sb, ob) in weeks.items(): print(f"{name:12} orders {ob - oa:+5d} share of change {(ob - oa) / drop:6.1%} conversion {oa / sa:.2%} -> {ob / sb:.2%}") # Step 2: walk the Android funnel before and after steps = ["sessions", "product views", "carts", "checkout started", "orders"] before = [40000, 28000, 8400, 2000, 1200] after = [41000, 28700, 8610, 2050, 820] for i in range(1, len(steps)): print(f"{steps[i - 1]} -> {steps[i]}: {before[i] / before[i - 1]:.1%} -> {after[i] / after[i - 1]:.1%}")

Output:Orders 3150 -> 2753 (-12.6%) Sessions 110000 -> 110000, conversion 2.86% -> 2.50% Android app orders -380 share of change 95.7% conversion 3.00% -> 2.00% iOS app orders +18 share of change -4.5% conversion 3.60% -> 3.60% Mobile web orders -20 share of change 5.0% conversion 2.00% -> 2.00% Desktop orders -15 share of change 3.8% conversion 3.00% -> 3.00% sessions -> product views: 70.0% -> 70.0% product views -> carts: 30.0% -> 30.0% carts -> checkout started: 23.8% -> 23.8% checkout started -> orders: 60.0% -> 40.0%

## Behavioural and stakeholder questions (Q27 to Q30)

- **Q27. Tell me about a project where your analysis changed a decision.** Use situation, task, action, result, the STAR method Amazon lists in its interview preparation. Give a real number from your own project and your exact role. See [data analytics projects with AI](https://www.masaischool.com/resources/projects/data-analytics-projects-with-ai) for portfolio ideas.
- **Q28. A stakeholder disagrees with your numbers. What do you do?** Ask what they expected and where it comes from, check definitions (period, filters, duplicates), reproduce both calculations, then agree one definition in writing. Stay curious, not defensive.
- **Q29. How do you present findings to a non-technical audience?** Lead with the recommendation, then one chart and the number behind it, then caveats. Use ranges, define terms once and leave out the tables you do not need.
- **Q30. What are your salary expectations?** Research ranges for your city and level before the call, using sourced figures such as the [data analyst salary in India](https://www.masaischool.com/resources/data-analyst-salary-in-india) page. Give a range tied to the scope of the role, and stay open to the whole package.

## A two-week preparation plan

| Days | Focus | Practice |
|---|---|---|
| 1 to 3 | SQL | Q1 to Q4 and the [SQL interview questions](https://www.masaischool.com/resources/sql-interview-questions); write five queries a day on a dataset from [free datasets](https://www.masaischool.com/resources/tools/free-datasets-for-projects) |
| 4 to 5 | Excel | Q5 to Q9; build a pivot dashboard from a public dataset |
| 6 to 7 | Statistics | Q15 to Q20; rerun the A/B script with your own numbers |
| 8 to 9 | Power BI | Q10 to Q14; rebuild one dashboard on a star schema |
| 10 to 12 | Cases | Q21 to Q26; solve two cases aloud against a timer, using the funnel script as a model |
| 13 to 14 | Behavioural and mocks | Q27 to Q30; record yourself and trim answers to two minutes. Add Python if the role asks for it |

Keep the [data analyst roadmap](https://www.masaischool.com/resources/data-analyst-roadmap) open to see where each skill sits, and note which questions you had to look up.

## Frequently asked questions

### What questions are asked in a data analyst interview?

Expect SQL (joins, grouping, window functions), Excel or Power BI tasks, statistics such as p-values and A/B tests, a business case like a metric drop, and behavioural questions. No company publishes its question bank, so this is our reading of common preparation topics.

### How many rounds are there in a data analyst interview in India?

It varies, and no employer publishes a standard. A common shape is a screening call, an SQL or Excel test, a case or technical interview and a hiring manager or HR round. Amazon's careers page says its process differs from role to role.

### How do I prepare for a data analyst interview in two weeks?

Split the time: three days of SQL, two of Excel, two of statistics, two of Power BI, three of cases and two of behavioural practice. The table above maps each block to questions. Practise aloud, because explaining your reasoning is graded as much as the answer.

### How do I explain a p-value in an interview?

Say that it measures how surprising the data would be if there were no real effect, and that it is not the probability the hypothesis is true. Then give an example, such as p = 0.044 for 520 versus 585 conversions, and add the confidence interval and the effect size.

### Do data analyst interviews ask Python?

Some do, especially when the posting mentions pandas or automation. Read the job description and prepare Python only if it is listed. If it is, work through the Python interview questions page, and know when you would choose SQL, Excel or Python for a task.

### What should a fresher say in a data analyst interview about projects?

Describe the question, the data source, the method, the result with a number and what you would do next. Be ready to explain every step and every chart. Two small projects explained well beat five listed without detail.

## Sources checked

- [Amazon: process differs from role to role; stages; STAR method listed](https://www.amazon.jobs/en/landing_pages/interviewing-at-amazon)
- [Masai post describing a four-round funnel](https://www.masaischool.com/blog/data-analyst-interview-questions-freshers-must-prepare-in-2026/)
- [ASA statement on p-values, 7 March 2016](https://www.amstat.org/asa/files/pdfs/p-valuestatement.pdf)
- [NIST confidence interval definition](https://www.itl.nist.gov/div898/handbook/prc/section1/prc14.htm)
- [Peeking inflates false positives, 26.1% example](https://www.evanmiller.org/how-not-to-run-an-ab-test.html)
- [Sample ratio mismatch in about 6% of Microsoft tests](https://www.microsoft.com/en-us/research/articles/diagnosing-sample-ratio-mismatch-in-a-b-testing/)
- [Star schema, measures stored at query time](https://learn.microsoft.com/en-us/power-bi/guidance/star-schema)
- [Import versus DirectQuery, Performance Analyzer](https://learn.microsoft.com/en-us/power-bi/connect-data/desktop-directquery-about)
- [Calculated columns recalculate on refresh](https://learn.microsoft.com/en-us/power-bi/transform-model/desktop-calculated-columns)
- [VLOOKUP first column and default approximate match](https://support.microsoft.com/en-us/office/vlookup-function-0bbc8083-26fe-4963-8ab8-93a18ad188a1)
- [XLOOKUP not available in Excel 2016 and 2019](https://support.microsoft.com/en-us/office/xlookup-function-b7fd680e-6d10-43e6-84f9-88eae8bf5929)
- [SUMIFS syntax](https://support.microsoft.com/en-us/office/sumifs-function-c9e748f5-7ea7-455d-9406-611cebce642b)
- [1,048,576 rows by 16,384 columns](https://support.microsoft.com/en-us/office/excel-specifications-and-limits-1672b34d-7043-467e-8e27-269d656771c3)
- [Power Query saves steps as a repeatable query](https://learn.microsoft.com/en-us/power-query/power-query-what-is-power-query)
