Data Science Interview Questions for Indian Companies 2026

The skill mix in Indian DS interviews depends heavily on role type and company stage.

Product analytics DS (Swiggy, Zomato, CRED, Meesho): heavy on SQL, A/B testing, and product metric interpretation. ML experience is nice-to-have, not required. ML engineering DS (Flipkart, Amazon, PhonePe): end-to-end ML pipeline: feature engineering, model selection, productionisation, model monitoring. Research DS (Microsoft Research India, Google DeepMind India, IBM Research): novel algorithm design, paper-level depth, typically requires a strong MS/PhD background. The mistake most Indian candidates make: preparing for an ML engineering role when applying for a product analytics role, or vice versa. Read the job description carefully for which skills are prioritised.

Statistics is the foundation tested in almost every Indian DS interview.

A/B testing: how to design a valid test (hypothesis, control/treatment groups, sample size calculation, runtime, p-value interpretation). Common question: 'Our A/B test shows p=0.04: do we ship the change?': the answer is not just yes; discuss practical significance, the multiple testing problem, and Bonferroni correction. Probability: conditional probability (Bayes' theorem), birthday problem, unfair coin: these appear in Flipkart and Amazon DS interviews. Distributions: normal, Poisson, binomial: know when each applies. Common question: 'If we see 10 customer support tickets per hour on average, what is the probability of seeing more than 15 in a given hour?': Poisson distribution.

ML questions vary widely by role: this covers what most Indian product company DS interviews test.

Bias-variance trade-off: high bias = underfitting (too simple model); high variance = overfitting (too complex model). Regularisation (L1/L2) as the lever. Cross-validation: k-fold cross-validation purpose and mechanics. Why use it instead of a single train/test split. Classification metrics: precision, recall, F1, AUC-ROC: when does recall matter more than precision? (fraud detection, medical diagnosis). Gradient Boosting: the dominant algorithm in Indian DS interviews: how XGBoost/LightGBM works conceptually, why it outperforms random forests in practice. Feature engineering: handling missing values (imputation strategies), categorical encoding (one-hot vs label vs target encoding), feature scaling (standardisation vs normalisation).

SQL is consistently the most tested technical skill in Indian data science interviews.

Expected SQL depth: window functions (ROW_NUMBER, RANK, LAG/LEAD), CTEs for complex queries, self-joins, handling NULLs in aggregations. Common DS SQL question: 'Find the top 3 products by revenue in each category, and show month-over-month revenue change.' Python for DS: Pandas (groupby, merge, pivot, handling missing data), NumPy (vectorised operations), Matplotlib/Seaborn for quick visualisations during take-home assignments. Scikit-learn: model training, cross-validation, GridSearchCV: you should be able to write a complete model pipeline (load data → feature engineering → train/CV → evaluate) from memory.

Product intuition questions are the differentiator between good and great DS candidates in Indian product companies.

Common format: 'Swiggy wants to reduce food delivery time: how would you approach this as a data scientist?' Expected answer structure: (1) Clarify what 'delivery time' means and what success looks like (metric definition). (2) Decompose into components (restaurant prep time, pickup wait, delivery route). (3) Identify data sources for each component. (4) Propose experiments or ML models for the highest-leverage component. (5) Discuss how to measure impact and handle edge cases. Business metric design: 'We want to improve user retention: what metrics would you track and what experiments would you run?': this tests whether you can think from first principles rather than just applying ML models.

Data science interview formats vary significantly across Indian companies. Flipkart and Meesho conduct a 90-minute online HackerEarth test with SQL, Python, and a machine learning conceptual section, followed by two technical rounds and one business case round. PhonePe emphasizes fraud detection scenarios and asks candidates to design features using transaction graph data. Zomato and Swiggy focus heavily on A/B testing design and causal inference for their growth data science roles. BFSI companies like HDFC, Paytm, and PolicyBazaar prioritize credit risk modeling, scorecard building using logistic regression, and regulatory compliance awareness such as RBI's FLDG norms. Knowing the domain context of the company you apply to is as important as knowing the algorithms.

In 2026, data science compensation in India spans a wide range depending on specialization and company tier. Freshers with strong Python and SQL skills can expect 8 to 12 LPA at mid-size startups and analytics firms. Data Scientists with 2 to 4 years of experience earn 18 to 28 LPA at product companies. Senior Data Scientists and ML Engineers at Tier 1 companies like Google, Flipkart, or PhonePe earn 35 to 55 LPA, with ESOPs forming 20 to 30 percent of total compensation. Specializations in NLP, recommendation systems, and real-time ML pipelines command a premium of 15 to 25 percent over generalist roles. Negotiation leverage increases sharply when you have a competing offer from another Tier 1 company.

Many Indian data science interviews now include a 24 to 72 hour take-home assignment before the final rounds. Common formats include a churn prediction task on anonymized e-commerce data, a time-series forecasting problem using sales data with seasonal Indian festival patterns, or an NLP classification task on product reviews in Hinglish. Evaluators look specifically at your EDA quality, feature engineering creativity, and whether you document trade-offs in model selection. Submitting a clean Jupyter notebook with markdown explanations, a confusion matrix, and a clear recommendation is more impressive than a slightly better AUC from an overfit model. Always include a section on how you would take the model to production as this signals engineering maturity beyond pure analysis.