A data analyst interview in India is rarely one conversation. It is usually a screening call, a SQL or Excel test, a case round on business judgement and a behavioural round, and candidates who prepare only for the technical part get filtered in the case round. Every question below has the answer an interviewer is listening for, and where a question hides a trap, the trap is named. They span SQL, Excel, Power BI, statistics and metric sense, because a real analyst round moves between them.
What each round is actually testing
Knowing which round you are in changes what a good answer looks like. Most Indian analyst processes take this shape.
| Round | What it tests | Usual format | What to prepare |
|---|---|---|---|
| Screening call | Basic definitions and whether your resume is real | 20 to 30 minutes with a recruiter or hiring manager | Two crisp lines on each project you have listed |
| SQL or Excel test | Whether you can query and clean data unaided | Take-home or shared screen, 45 to 60 minutes | Joins, aggregation, window functions, lookups, pivot tables |
| Case or metric round | Judgement, metric definitions and diagnostics | Open-ended problem with the hiring manager | Frameworks for a metric drop and success measurement |
| Portfolio review | Communication and design choices | You present your own dashboard or analysis | One project you can defend end to end |
| Behavioural | Ownership and stakeholder handling | Manager or skip-level round | Two or three real stories with numbers attached |
Fresher level: the screening round
These check that you understand the job rather than just the tools.
What does a data analyst actually do day to day?
Most of the day goes into pulling data with SQL, cleaning and reconciling it, and turning it into a chart, a table or a short written answer for someone making a decision. Less goes into maintaining dashboards, and less still into genuinely new analysis. Saying cleaning and reconciliation take the largest share is more credible than describing the job as insight generation.
Which tools should a data analyst know?
SQL for retrieving data, Excel for the last mile, and a visualisation tool, usually Power BI or Tableau, for recurring reporting. Python helps for larger cleaning jobs and is expected in some product roles. Statistics sits underneath all of it, because the tools will happily compute a number that means nothing.
What are the steps in cleaning a dataset?
Profile it first: row counts, distinct counts, ranges and the share of missing values per column. Then fix structural issues such as wrong data types and inconsistent category spellings, and handle duplicates, missing values and outliers with a documented rule. Finish by validating totals against a source the business already trusts, because an analysis that does not reconcile to the finance number gets rejected regardless of method.
What is the difference between mean, median and mode, and when does the median matter more?
The mean is the arithmetic average, the median the middle value when sorted, the mode the most frequent value. The median is the honest choice whenever the distribution has a long tail, which describes most business data: salaries, order values, session durations. Reporting mean order value on a dataset containing a few bulk orders tells the reader almost nothing about a typical customer.
What is a KPI and how is it different from a metric?
A metric is any number you can measure. A KPI is the small set of metrics an organisation has agreed to be judged on, so it needs an owner, a target and a review cadence. If everything on the dashboard is a KPI, none of them are.
How do you choose a chart type?
Match the chart to the comparison the reader needs to make: bars for categories, lines for change over time, scatter for two continuous variables, a plain number for a headline figure. Pie charts are defensible only for two or three parts of a whole. Avoid dual axes, because rescaling them lets you imply any relationship you like.
Intermediate level: the technical round
Most offers are decided here, with a mix of query writing, spreadsheet reasoning and statistics.
Explain the difference between an INNER JOIN and a LEFT JOIN.
An INNER JOIN keeps only rows with a match on both sides. A LEFT JOIN keeps every row from the left table, filling right-hand columns with nulls where no match exists. Its practical use is finding what is missing: join customers to orders, filter to rows where the order key is null, and you have every customer who never bought.
When would you use a window function instead of GROUP BY?
Use GROUP BY when you want one summary row per group. Use a window function when you need each original row kept alongside a group-level calculation, such as a running total, a rank within a city, or an order shown next to that customer’s average. They are also the cleanest route to month-on-month change using LAG.
Why do analysts prefer XLOOKUP or INDEX with MATCH over VLOOKUP?
VLOOKUP cannot look left of its search column, and its column index is a hard number that breaks when somebody inserts a column. XLOOKUP defaults to an exact match, searches and returns from independent ranges, and takes a not-found argument. INDEX with MATCH does the same and works in older Excel versions, which still matter in many Indian offices.
In Power BI, when do you write a measure and when a calculated column?
Write a measure by default, since it is evaluated at query time under the visual’s filters and stores nothing. Use a calculated column only when the value must exist per row before filtering, for example to slice by it or put it on an axis. The trap is memory, since calculated columns are stored and compress poorly when they hold many distinct values.
How do you handle missing values?
First find out why they are missing, because a null meaning zero, one meaning not applicable and one meaning the pipeline broke deserve different treatment. Then choose between dropping the rows, imputing with a median, or carrying the last known value forward, and document the choice on the output. Never impute silently, because a reader assumes a filled cell is observed data.
How do you detect and treat outliers?
Look at them with a histogram or box plot before applying any rule. The two standard detection methods are the interquartile range rule and a z-score threshold, and the choice depends on whether the distribution is skewed. Treatment depends on cause: a data-entry error gets fixed or dropped, whereas a genuine bulk order gets kept and reported separately, because removing it hides real revenue.
What is the difference between correlation and causation, and how do you explain it to a stakeholder?
Correlation says two things move together; causation says one produces the other. The gap is usually a third variable or the effect running the other way. With a stakeholder, give the concrete alternative rather than the abstract principle: users of the referral feature spend more, and the plausible reason is that heavy users refer, rather than referring making people spend.
What is a p-value in plain language?
It is the probability of seeing a result at least as extreme as the one you observed, assuming there is genuinely no effect. A small p-value means the data would be surprising if nothing were happening, so we treat it as evidence against the no-effect assumption. It is not the probability that your hypothesis is true, and saying so is what interviewers are checking.
What is the difference between a Type I and a Type II error?
A Type I error is a false positive: you conclude there is an effect when there is none. A Type II error is a false negative, a real effect the test missed, usually because the sample was too small. Which is worse is a business question, so name the trade-off in context, such as fraud detection where a miss costs more than a wrongly flagged transaction.
What is selection bias? Give a business example.
Selection bias is when the data you analysed does not represent the population you want to conclude about. A common example is running a satisfaction survey inside the app: you never hear from users who uninstalled, so satisfaction looks healthier than it is. The fix is sampling from the full population, or being explicit about who the conclusion applies to.
How would you design an A/B test for a change to the checkout page?
Define the primary metric before you start, usually completed orders per visitor rather than clicks, plus a guardrail metric such as average order value. Randomise at the visitor level, size the sample for the smallest effect worth detecting, and fix the duration in advance to cover a full weekly cycle. Do not stop when the result first looks good, because repeated peeking inflates the false-positive rate.
Advanced level: business sense and stakeholder rounds
These separate candidates with identical technical scores.
Daily orders dropped 20% overnight. How do you investigate?
Confirm the number is real first: check whether a pipeline failed, a tracking change shipped, or the drop appears in only one source. Then cut the metric by dimension one at a time, across platform, city, channel, new versus returning users and payment method, because a genuine drop is usually concentrated rather than uniform. Line the timing up against releases, campaigns, festivals and outages, and report what you ruled out.
How would you measure whether a new feature succeeded?
Start from what the feature was supposed to change, then pick one primary metric that moves only if that happened. Add a counter-metric to catch damage elsewhere, such as a faster checkout raising refunds. State the time window and comparison group up front, because a feature judged on its launch week always looks better than it is.
A sales head asks you for a dashboard. What do you ask before building it?
What decision will this change, how often will you look at it, and what will you do differently when a number moves. Then confirm definitions, since revenue can mean gross, net of returns or booked. Ask who else will see it, because that sets row-level security and the level of aggregation.
Two reports show different revenue for the same month. How do you resolve it?
Do not adjust either number until you know why they differ. Compare the grain, the date field used, the filters, the tax handling and the cut-off timestamp, because most mismatches come from one of those rather than a calculation error. Then write the definition down where both teams can see it, or it resurfaces next quarter.
How do you explain a statistical result to a non-technical stakeholder?
Lead with the decision implication in one sentence, then the number, then how confident you are in it. Replace the vocabulary with business language: say the improvement is plausibly between 2% and 9% rather than quoting a confidence interval as such. Keep the method available for anyone who asks, but do not open with it.
Tell me about a time your analysis turned out to be wrong.
Answer with a real case, name the specific error such as a join that duplicated rows or a date filter that excluded the last day of the month, and say how it was caught and what you changed. A candidate who claims this has never happened is either inexperienced or not checking their work. The habit you built afterwards, such as always reconciling a total against a trusted source, is what interviewers remember.
Preparing without burning three months
Pick one project and take it end to end. Pull data with SQL, clean it, build a small model, put it in Power BI and write a one-page conclusion someone could act on. That single project answers a surprising number of the questions above and gives your portfolio round something to talk about.
Practise the case round aloud with another person, because the technical answers are learnable alone while the metric-drop and dashboard-scoping questions need someone interrupting you with follow-ups. And name your weak area: saying you are strong on SQL and reporting but have limited experience designing experiments beats a confident wrong answer, and it survives a reference check.
Frequently asked questions
How many rounds does a data analyst interview usually have in India?
Three to five is typical: a screening call, a SQL or Excel test, a case or hiring-manager round, and often a behavioural round. Service companies and staffing firms run fewer and lean on tool questions, while product companies weight the case round heavily. The process usually runs one to three weeks.
What is the most important skill in a data analyst interview?
SQL, because it appears in almost every process and is the easiest thing to test objectively. Business judgement decides between two candidates who both cleared the SQL round, and it is what most freshers neglect.
Do freshers get asked case questions?
Yes, in a simplified form. A fresher is more likely to be asked how they would measure whether a promotion worked than to design a full experiment. Interviewers are checking whether you think about the business behind the numbers, and accept a structured answer even when incomplete.
Should I take a Python question seriously if the job description says SQL and Excel?
Prepare enough Python to read code and describe a cleaning workflow with pandas, because interviewers sometimes ask even when the role does not require it. Depth is not expected unless the description mentions it. Spending preparation time on Python instead of SQL is the commoner mistake.
How do I answer when I do not know something?
Say so, then say how you would find out or reason from what you do know. Interviewers accept gaps and rarely accept bluffing, since a confident wrong answer warns them how you will behave with a stakeholder. Ask a clarifying question when the problem is genuinely ambiguous, because that is itself a tested behaviour.
What salary should a fresher data analyst expect in India?
Entry-level offers commonly sit in the 4 to 7 LPA range depending on city, sector and how technical the role is, with product companies at the higher end. SkilloVilla alumni report packages between 4 and 15 LPA with a median of 9.5 LPA, and outcomes depend heavily on prior background. Check several sources before negotiating.
Prepare with mentors who have taken these interviews
Reading answers builds recognition. Being asked a follow-up you did not expect, and having a working analyst explain why your answer would not have landed, builds the skill that gets tested.
SkilloVilla’s Data Analytics with Python track at ₹71,999, currently ₹58,999, covers SQL, Excel, Power BI, Python and statistics with live classes, 1:1 mentorship from working analysts and interview preparation, plus placement support and alumni at Accenture, PwC, Flipkart, Razorpay and Swiggy. Scholarships are available depending on your profile.
Fees and ratings last checked August 2026; confirm current numbers with the provider before enrolling.
Still choosing a programme? Read data analyst qualifications in India, compare prices in our data analyst course fees breakdown, and see what support you are buying in placement assistance explained.
