Flipkart Data Scientist Interview: Questions, Experience & Prep (2026)

Flipkart is one of India's largest e-commerce platforms, and its Data Science team works on problems at a scale few Indian companies match: recommendation systems, demand forecasting, fraud detection, dynamic pricing, and supply chain optimisation. As of July 2026, knok's job radar shows 37 open Data Scientist roles at Flipkart.

The interview process typically spans several rounds. Candidates report it covers statistics and probability, machine learning depth, SQL and data manipulation, product sense, and a behavioural conversation. Expect the whole process to take a few weeks from first contact to offer.

Salary ranges from the knok job radar:

Bangalore leads hiring with 166 active Data Scientist roles as of July 2026. Delhi (46 roles) and Hyderabad (27 roles) are the next busiest cities.

These questions come up most often, based on what candidates report from Flipkart Data Scientist interviews. They skew toward applied ML, e-commerce business problems, and SQL.

1. How would you design a product recommendation system for Flipkart's homepage?
2. How would you build a demand forecasting model for Flipkart's Big Billion Days sale?
3. You run an A/B test on the search ranking. Clicks go up but conversions drop. How do you investigate and decide?
4. How would you detect fraudulent sellers or fake product reviews at scale?
5. Walk me through how you would predict customer churn for Flipkart Plus subscribers.
6. How would you design a dynamic pricing algorithm for categories with high competition?
7. How would you measure whether a new app feature is actually working?
8. A model you deployed to production has started degrading. How do you debug it?
9. Write a SQL query to find the top-selling product in each category over the last quarter.
10. How would you use data to reduce last-mile delivery delays?
11. Tell me about a time you used data to challenge or change a business decision.
12. How do you handle severely imbalanced classes in a fraud detection model?

Q: Tell me about a time you used data to change a business decision.

*Situation:* At my previous company, the product team planned to discontinue a category they believed was underperforming, based on raw revenue numbers alone.

*Task:* I was asked to validate their analysis before the decision went to leadership.

*Action:* I pulled transaction data and found buyers in this category had a disproportionately high repeat purchase rate. I built a simple customer lifetime value model showing these buyers also spent more across other categories. I presented a visual comparison of short-term revenue against long-term customer value to the product head.

*Result:* The team paused the discontinuation, ran a focused experiment instead, and the category was retained. The CLV model became a standard tool the team now uses for similar decisions.

---

Q: You ran an A/B test and got conflicting signals across metrics. How did you handle it?

*Situation:* I was testing a new sort-order algorithm for a product listing page. Clicks on featured products went up, but overall cart adds dropped.

*Task:* I needed to make a clear recommendation on whether to ship the change.

*Action:* I segmented the data by device type, user cohort, and product category to isolate where the drop was coming from. I found users on lower-end devices were seeing slower page loads with the new layout, which was hurting cart adds for that segment. I flagged this to engineering and proposed a follow-up test after a performance fix.

*Result:* The follow-up test showed consistent improvement across all key metrics. We shipped the fix and the new algorithm together, avoiding a rollout that would have hurt a large segment of users.

---

Q: Describe a time you identified and fixed a data quality issue in a production model.

*Situation:* A fraud detection model I owned started flagging fewer transactions than expected, shortly after a backend pipeline change.

*Task:* I needed to find the root cause and restore reliability without rolling back the pipeline, which other teams depended on.

*Action:* I traced the issue to a feature engineering step that assumed a specific date format, which the new pipeline had changed. I rewrote the parsing logic to handle both formats, added a validation layer to the feature pipeline, and set up monitoring alerts for feature distribution drift.

*Result:* Model performance returned to baseline the next day. The monitoring layer has since caught other silent data issues before they could affect model output.

For ML system design questions (recommendations, forecasting, fraud detection):

Start by framing the business problem: what are you predicting, and why does it matter to Flipkart? Then walk through data availability, your modelling approach and why (baseline first, then complexity), your evaluation metrics (distinguish business metrics from model metrics), and finally deployment and monitoring. Interviewers want end-to-end thinking, not just algorithm knowledge.

For A/B testing and metrics questions:

Clarify the hypothesis and the primary metric first. Then discuss guardrail metrics (what you cannot let slip), segmentation (is the effect consistent across user groups?), and the decision rule. Candidates who mention novelty effects, network effects, or long-run holdout analysis signal stronger depth.

For SQL questions:

Think out loud. State what the query needs to return, identify the joins and groupings needed, then write the SQL. Window functions such as RANK, ROW_NUMBER, and LAG come up often. Practice writing clean, readable queries, not just ones that technically work.

For behavioural questions:

Use STAR: Situation (brief context), Task (your specific role), Action (what YOU did, not the team), Result (measurable or observable outcome). Keep Situation and Task short. Spend most of your time on Action and Result.

Domain connection. Flipkart interviewers consistently look for candidates who link machine learning to e-commerce business outcomes. Knowing gradient boosting is not enough. Know why you would use it for demand forecasting versus a simpler baseline, and what getting it wrong costs the business.

Scale awareness. Flipkart operates at a scale that changes many design decisions. Candidates who address computation cost, data freshness, and real-time versus batch tradeoffs stand out over those who give textbook answers.

End-to-end ownership. Be ready to discuss shipping models to production, monitoring, retraining cadence, and what you do when a model degrades. Notebook-only experience is not sufficient.

Clear communication. Data Science at Flipkart involves working with product managers, engineers, and business stakeholders. Candidates who explain their reasoning step by step, without jargon, are rated higher in the process.

Intellectual honesty. If you don't know something, reason through it aloud. Candidates report that interviewers respond well to 'I haven't worked with this directly, but here is how I would approach it' rather than bluffing.

Build your foundations first.

Review statistics and probability: Bayes' theorem, common distributions, hypothesis testing, and confidence intervals. Practice SQL with window functions and multi-level aggregations on large tables. These topics appear in early screening rounds.

Go deep on topics Flipkart cares about.

Focus on tree-based models (XGBoost, LightGBM), recommendation systems (collaborative filtering, matrix factorisation, two-tower models), time series forecasting, and anomaly detection for fraud. For each topic, be ready to explain model choice, tradeoffs, and evaluation metrics.

Practice e-commerce case studies.

Work through the questions listed in this guide. For each system design question, sketch a rough architecture covering data ingestion, feature engineering, modelling, and serving. Get comfortable with Flipkart-specific contexts: Big Billion Days, Flipkart Plus, seller marketplace, and last-mile logistics.

Prepare your behavioural stories.

Have several strong STAR stories ready: one where you changed a decision with data, one where you fixed a production issue, and one where you worked cross-functionally. Rehearse these out loud, not just in your head.

Run timed mock sessions.

Candidates report that practicing SQL and ML questions under a time limit significantly reduces nerves in actual rounds. Do at least a few of these before your first interview.

While you prep, knok checks 150+ job sites nightly, applies to jobs matching your resume, and messages HR for you, so you don't miss active Flipkart openings.

Jumping to algorithms before framing the problem. The most common mistake candidates report is naming a model in the first sentence of a system design answer. Interviewers expect you to clarify the business goal, the data available, and what 'good' looks like before any model is mentioned.

Ignoring data quality and pipeline issues. Many candidates describe a perfect world where clean data arrives on schedule. Flipkart interviews often probe what happens when data is late, missing, or drifts. Always mention validation and monitoring in your answer.

Treating all A/B test metrics equally. Saying 'clicks went up, so it is a win' is a red flag. Be explicit about primary metrics, guardrail metrics, and what trade-offs are acceptable to the business.

Generic STAR answers. Answers like 'I improved model accuracy' without context or business impact sound hollow. Always anchor the result to something the business cared about.

Underestimating the SQL round. Candidates sometimes treat SQL as a formality and underperform. Flipkart's SQL questions can involve multi-level aggregations and window functions. Practice these under time pressure.

Not asking clarifying questions. Interviewers expect you to ask about scale, constraints, and success criteria before diving in. Going straight to an answer without any clarification signals poor problem-solving instincts.