Data & Analytics Interview

Data, AI and enterprise-platform hiring in India typically runs an HR or recruiter screen, a hiring-manager round on past projects, one or two technical rounds (a live SQL or case discussion for analytics, a modelling and systems discussion for ML, a configuration and scenario walkthrough for SAP, Salesforce, ServiceNow and Dynamics), and a behavioural or fitment round, spread over one to three weeks. Many IT-services and GCC loops add a client interview before the offer. ExamPilot compresses this into one 17-minute panel: HR screen, hiring-manager round on what you actually built and the business result, a role-specific technical round led by a lead or principal, a STAR behavioural round and a closing on expectations. Rounds are pass/fail in real companies; here every round is scored on role knowledge, problem solving, communication and attitude, with specific feedback on the trade-offs you did and did not mention.

1. 1

   SQL

   Say aloud the SQL that pulls every customer's second-latest order from an orders table. Then convince me your output is correct.

   What a strong answer covers

   Use ROW_NUMBER partitioned by customer and ordered by order date descending, then filter where the rank equals two. Discuss ties and whether DENSE_RANK fits better. Verify by checking a few customers manually and comparing counts. Avoid self-joins that break on ties or skipping verification.

2. 2

   Joins

   After you join orders to a promotions table, total revenue in your report suddenly doubles. What has probably happened, and how do you fix it?

   What a strong answer covers

   Explain fan-out: one order matched several promotion rows, duplicating revenue. Diagnose by checking join keys and row counts before and after the join, then fix with the correct grain, pre-aggregation or a deduplicated bridge. Avoid patching it with DISTINCT on revenue without understanding the cause.

3. 3

   KPI debugging

   Weekly active users on the dashboard dropped twenty percent overnight. Walk me through how you work out whether it is real.

   What a strong answer covers

   First check the pipeline: late or failed loads, schema changes, tracking or app-release issues. Then compare with independent sources like server logs or orders, segment by platform and region, and only then look for real business causes. Communicate early with a confidence level. Avoid announcing a crisis before ruling out data problems.

4. 4

   Data modelling

   Design a data model for a food-delivery company that needs daily reports on orders, delivery times and restaurant ratings.

   What a strong answer covers

   Propose a star schema with an orders fact table at order grain, dimensions for customer, restaurant, area and date, and a separate fact for ratings if grain differs. Explain slowly changing dimensions and partitioning by date. Avoid one giant flat table without stating the grain.

5. 5

   Pipeline design

   As a data engineer, would you build a nightly batch pipeline or a streaming pipeline for a fraud dashboard at a payments company? Defend your choice.

   What a strong answer covers

   Match latency to the business need: fraud detection usually needs near real-time, so streaming with Kafka or a managed equivalent, while reporting aggregates can stay batch. Discuss cost, complexity, exactly-once processing and backfills. Avoid choosing streaming just because it sounds advanced.

6. 6

   Experiments

   The checkout team split traffic between the old and a redesigned pay button, and after two days the new one is up three percent. Would you call it a win?

   What a strong answer covers

   Check whether the test reached its pre-planned size and significance, cover full weekly cycles, look for novelty effects and watch guardrails such as refunds and payment failures. Recommend running to the planned length. Avoid declaring a winner from early peeking or ignoring segments where the change hurt.

7. 7

   BI performance

   For a BI developer: a Power BI report takes forty seconds to load and the business head is complaining. Where would you start?

   What a strong answer covers

   Check the data model for unnecessary columns and bidirectional relationships, review heavy DAX measures, use the Performance Analyzer, consider aggregations or import mode instead of DirectQuery, and reduce visuals per page. Measure before and after. Avoid buying more capacity as the first answer.

8. 8

   Business impact

   Tell me about a dashboard or analysis you built that actually changed a business decision.

   What a strong answer covers

   A STAR answer: the question the business had, your analysis or dashboard, the insight, the decision it led to and the measurable result. Show you understand the business, not just the tools. Avoid describing a dashboard nobody used or listing charts without an outcome.

9. 9

   Mistakes

   You shipped a report with a wrong number that a VP had already presented. What did you do?

   What a strong answer covers

   Own it quickly, inform your manager and the VP with the corrected number and its impact, explain the cause briefly, and add a check to prevent recurrence. Show calm and integrity. Avoid hiding the error or quietly fixing it without telling anyone.

• English
• Hinglish (Hindi + English)Hinglish