Question 1

What is the difference between WHERE and HAVING?

Accepted Answer

WHERE filters individual rows before aggregation; HAVING filters groups after GROUP BY has aggregated them. You cannot use an aggregate such as SUM() in WHERE, but you can in HAVING. Example: WHERE order_date >= '2025-01-01' GROUP BY customer_id HAVING SUM(amount) > 1000.

Question 2

Explain INNER, LEFT, RIGHT and FULL OUTER JOIN.

Accepted Answer

INNER returns only matching rows in both tables. LEFT returns all rows from the left table plus matches (NULLs where none). RIGHT is the mirror image. FULL OUTER returns all rows from both sides with NULLs where there is no match. In reporting, LEFT JOIN is the safe default so you don't silently drop customers with no orders.

Question 3

How do you find the second highest salary per department?

Accepted Answer

Use a window function: SELECT * FROM (SELECT emp_id, dept_id, salary, DENSE_RANK() OVER (PARTITION BY dept_id ORDER BY salary DESC) rnk FROM employees) t WHERE rnk = 2. DENSE_RANK handles ties by giving equal salaries the same rank; ROW_NUMBER would arbitrarily pick one.

Question 4

How do you handle missing values in a sales dataset?

Accepted Answer

First diagnose: how many, which columns, and is the missingness random or structural (e.g. discount blank means zero)? Then choose: drop rows only if the loss is negligible, fill structural blanks with a documented default, impute numeric columns with median (robust to outliers) or a group median, and for time series use forward fill. Always add a flag column so downstream models know a value was imputed.

Question 5

Difference between loc and iloc?

Accepted Answer

loc is label-based (df.loc[df.country=='Kuwait','revenue']); iloc is integer-position based (df.iloc[0:5, 2]). Mixing them is the most common pandas bug after a reset_index.

Question 6

Explain merge vs join vs concat in pandas.

Accepted Answer

merge is a SQL-style join on columns or indexes with how='inner|left|right|outer'. join is a convenience wrapper that joins on index by default. concat stacks frames vertically (axis=0) or side by side (axis=1) without key matching.

Question 7

When do you use a measure instead of a calculated column?

Accepted Answer

A calculated column is computed row by row at refresh and stored in the model — use it when you need a value to slice, filter or relate on. A measure is computed at query time in the filter context — use it for aggregations such as totals, ratios and time intelligence. Measures keep the model small and respond correctly to slicers.

Question 8

Explain star schema and why it matters in Power BI.

Accepted Answer

One central fact table of events joined to dimension tables (date, product, customer). It gives simple one-to-many single-direction relationships, faster VertiPaq compression, and predictable DAX. Snowflaking or joining fact to fact leads to ambiguous filter paths and slow reports.

Question 9

What does CALCULATE do?

Accepted Answer

CALCULATE evaluates an expression in a modified filter context. It is the only function that can override existing filters, e.g. CALCULATE([Sales], ALL(Date)) for a grand total or CALCULATE([Sales], Product[Category]="Electronics"). Understanding context transition — row context becoming filter context inside CALCULATE — is the core DAX concept.

Question 10

XLOOKUP vs VLOOKUP vs INDEX-MATCH?

Accepted Answer

VLOOKUP only searches right from the first column and breaks when columns move. INDEX-MATCH looks in any direction and is column-insertion safe. XLOOKUP replaces both with exact match by default, a built-in if-not-found argument and reverse search. Use XLOOKUP where available, INDEX-MATCH for older files.

Question 11

How would you build a dynamic KPI summary?

Accepted Answer

Load data with Power Query, create a table, then use SUMIFS/COUNTIFS driven by cell-linked slicer values or a PivotTable with GETPIVOTDATA. Add conditional formatting for variance and a spark-line trend column. Keep raw data, calculations and presentation on separate sheets.

Question 12

What is Power Query used for?

Accepted Answer

Repeatable ETL inside Excel: connect to files, folders, databases and APIs; unpivot, split, merge and clean; then load to the sheet or the data model. Every step is recorded so next month you press Refresh instead of redoing the work.

Question 13

What does a p-value actually tell a business stakeholder?

Accepted Answer

It is the probability of seeing a result at least this extreme if there were genuinely no effect. A small p-value means the data would be surprising under 'no difference', so we act as if a difference exists. It does not tell you the size of the effect or the probability that your hypothesis is true — always report the effect size and confidence interval alongside it.

Question 14

Explain Type I and Type II errors with a business example.

Accepted Answer

Type I (false positive): we launch a new checkout that actually does nothing, wasting development budget. Type II (false negative): we discard a change that would have increased conversion. The significance level controls Type I; sample size and effect size control Type II via statistical power.

Question 15

Mean, median or mode — which for salary reporting?

Accepted Answer

Median, because salary distributions are right-skewed and a few executive packages drag the mean up. Report the median with quartiles; use the mean only alongside a note on skew.

Question 16

How do you detect and correct data leakage?

Accepted Answer

Leakage is when information unavailable at prediction time enters training — e.g. a 'refund_date' column when predicting refunds, or scaling fitted on the full dataset before the split. Detect it by suspiciously high validation scores and by auditing feature timestamps. Fix by splitting first, fitting all transformations inside a pipeline on the training fold only, and using time-based splits for temporal data.

Question 17

Explain the bias-variance trade-off.

Accepted Answer

High bias (underfitting) means the model is too simple and misses real structure; high variance (overfitting) means it memorises noise and fails on new data. Manage with model complexity, regularisation, more data, and cross-validation to measure generalisation rather than training fit.

Question 18

Your model has 97% accuracy on a fraud dataset. Is it good?

Accepted Answer

Almost certainly not — if 3% of transactions are fraud, predicting 'not fraud' every time scores 97%. Use precision, recall, F1, PR-AUC and a confusion matrix, choose the threshold from the business cost of a false negative versus a false positive, and consider resampling or class weights.

Question 19

ETL vs ELT — which and when?

Accepted Answer

ETL transforms before loading, which suits limited warehouse compute and strict pre-load governance. ELT loads raw data first and transforms inside a modern warehouse (Snowflake, BigQuery, Databricks) where compute is elastic — this is the default today because raw data stays replayable.

Question 20

What is idempotency in a pipeline and how do you achieve it?

Accepted Answer

Running the same job twice produces the same result. Achieve it with deterministic partition overwrites (delete-then-insert by date partition), MERGE/upsert on a natural key, and avoiding blind appends. It is what makes safe backfills and retries possible.

Question 21

Explain slowly changing dimensions.

Accepted Answer

SCD Type 1 overwrites the old value (no history). Type 2 inserts a new row with valid_from/valid_to and an is_current flag, preserving history — the standard for customer or price dimensions. Type 3 keeps a previous-value column for a single change.

Question 22

Tell me about a time your analysis changed a decision.

Accepted Answer

STAR answer: Situation — marketing planned to double spend on a channel. Task — I was asked to validate performance. Action — I rebuilt attribution from raw event data, found 40% of conversions were double-counted across channels, and presented a corrected dashboard. Result — the budget shifted to two channels with genuine incremental return and cost per acquisition fell 18%.

Question 23

Describe a time you made a mistake in an analysis.

Accepted Answer

Own it fast and show the system fix: I shipped a report with a mis-joined table that inflated revenue by 6%. I flagged it myself within hours, reissued corrected numbers with a note, and added a row-count reconciliation check to every report I publish. Interviewers reward transparency and prevention over perfection.

Question 24

How do you handle a stakeholder who rejects your findings?

Accepted Answer

Separate the data from the interpretation. Walk through the source and method, invite them to challenge assumptions, quantify uncertainty honestly, and offer a follow-up test. If they hold domain knowledge you lack, incorporate it — often they are pointing at a real data gap.

Question 25

Why analytics, and why this company in the Gulf market?

Accepted Answer

Connect three things: a genuine origin story (a problem you solved with data), evidence of commitment (projects, certification, portfolio), and specific research on the company — their market, a product, and a metric you would want to improve in the first 90 days.

Question 26

What are your salary expectations?

Accepted Answer

Give a researched range, not a single number, and anchor it to the market: 'Based on Kuwait market data for analysts with dashboard and SQL delivery experience, I'm targeting KWD 550–700 per month, and I'm flexible depending on the full package including allowances and visa support.' Never state a number before you understand the package structure.

Question 27

What should I know about Gulf package structures?

Accepted Answer

Offers are often split into basic salary plus housing, transport and sometimes education allowances, with annual leave, an air ticket and end-of-service gratuity. Compare total monthly cash and the gratuity basis, not just the headline. Confirm visa type, transferability and probation length in writing.

Question 28

Revenue dropped 12% last quarter. How do you investigate?

Accepted Answer

Decompose revenue = customers × orders per customer × average order value, then segment each component by region, channel, product category and new vs returning. Check for data issues first (missing store feed, tracking break). Compare with seasonality and last year. Isolate whether the drop is broad or concentrated, then quantify each contributing factor and rank fixes by impact.

Question 29

A dashboard's numbers don't match finance. What do you do?

Accepted Answer

Agree the exact definition and time window, then reconcile step by step: source table, filters, currency conversion, returns and cancellations, timezone and cut-off. Produce a reconciliation table showing each adjustment. Publish an agreed metric definition afterwards so it does not recur.

Question 30

How would you measure the success of a new mobile app feature?

Accepted Answer

Define the behaviour change you expect, pick one primary metric (e.g. weekly active users completing the flow), add guardrails (crash rate, support tickets, revenue), instrument events before launch, roll out to a randomised subset, and evaluate with pre-registered analysis after a full cycle.

Question 31

What is the difference between a calculated column and a measure?

Accepted Answer

A calculated column is computed row by row at refresh and stored in the model, increasing size. A measure is calculated at query time in the current filter context and is not stored. Use measures for aggregations shown in visuals; use columns only when you need to slice or filter by the value.

Question 32

Explain filter context and row context.

Accepted Answer

Row context is the current row during iteration (calculated columns, SUMX). Filter context is the set of filters applied by slicers, visuals and CALCULATE. CALCULATE performs context transition, turning row context into an equivalent filter context.

Question 33

Why is a star schema recommended in Power BI?

Accepted Answer

Fact tables hold numeric events and dimension tables hold descriptive attributes, joined by one-to-many relationships. This keeps DAX simple, filters propagate predictably, compression is better and reports are faster than with flat or snowflaked tables.

Question 34

What is the role of a Business Analyst?

Accepted Answer

A BA understands business problems, gathers and documents requirements, maps processes, defines success metrics and works with stakeholders and technical teams so the delivered solution actually solves the problem.

Question 35

Functional vs non-functional requirements?

Accepted Answer

Functional requirements describe what the system must do (e.g. generate an invoice). Non-functional requirements describe how well it must do it — performance, security, availability, usability.

Question 36

How do you write a good user story?

Accepted Answer

As a , I want , so that , plus clear acceptance criteria (Given/When/Then). Keep it small, testable and valuable — the INVEST criteria.

Question 37

What is retrieval-augmented generation (RAG)?

Accepted Answer

RAG retrieves relevant documents from a knowledge base (usually via vector search) and passes them into the LLM prompt, so answers are grounded in your data, more current and less likely to hallucinate — without retraining the model.

Question 38

What are embeddings?

Accepted Answer

Dense numeric vectors representing the meaning of text (or images). Similar meanings sit close together, which enables semantic search, clustering and recommendation.

Question 39

How do you reduce hallucinations?

Accepted Answer

Ground with RAG, instruct the model to answer only from context and say 'I don't know', lower temperature, use structured outputs, add citation requirements and evaluate with a test set of known answers.

87 interview questions with full model answers

Mock Interviews

Coding Tests

AI Feedback

Difficulty Levels

SQL

Python

Power BI

Excel

Statistics

Machine Learning

Data Engineering

Behavioural

HR & Salary

Case Studies

Power BI & DAX

Business Analysis

Generative AI & LLMs

Best interview preparation videos — English and Hindi/Urdu

Data Analyst Interview Questions and Answers (full walkthrough)

Top SQL Interview Questions with live query solving

SQL Interview Questions in Hindi — practice set

Data Analyst Interview Preparation (Hindi)

Power BI Interview Questions and Answers

Excel Interview Questions for Data Analysts

Python Interview Questions for Data Analysts & Scientists

Machine Learning Interview Questions (Hindi/English)

Statistics Interview Questions for Data Science

Data Engineering Interview Questions (Spark, Kafka, Airflow)

Tell Me About Yourself — best answer framework

Salary Negotiation for Gulf & remote jobs

Want a mock interview with a mentor?

Rehearse until it feels routine

Study resources for every interview

Free courses, docs & certifications

Top-rated YouTube videos

Ready? Apply on the top job sites

Need Help Choosing the Right Career?