We are looking for a Senior Data Scientist to build predictive and diagnostic models for a large retail and direct-selling client. You will work with substantial structured sales, customer, marketing and CRM datasets to explain business performance, predict customer behaviour and help managers make better decisions.

You should enjoy investigating data as much as building models: questioning definitions, uncovering unexpected patterns, testing competing explanations and determining whether an apparent finding survives rigorous validation. This role requires experience with imperfect business data, statistical reasoning and production delivery in Databricks.

Key responsibilities
- Lead data-quality assessment and exploratory data analysis across customer acquisition, orders, sales and marketing datasets.
- Investigate missingness, duplicate records, inconsistent identifiers, join inflation, incomplete customer histories, delayed data and changing business definitions.
- Quantify how these issues affect model feasibility and accuracy.
- Use statistical analysis to distinguish meaningful patterns from sampling noise, population changes, measurement artefacts and confounding.
- Build, benchmark and validate models for purchase progression, churn, customer value.
**Essential knowledge and experience**
- Robust SQL and Python: Experience analysing large relational datasets, verifying join cardinality and data grain
- Applied statistics and EDA: Distributions, conditional relationships, confidence intervals, bootstrap methods, hypothesis testing, effect sizes, multiple comparisons and selection bias. Understand why millions of rows may still provide limited independent evidence.
- Customer and cohort analysis:



Conversion and retention curves, recency/frequency/value features, repeat-purchase behaviour, outcome maturity and changing customer populations.
- Predictive modeling: Logistic regression and gradient-boosted trees such as LightGBM or XGBoost; thoughtful feature engineering, regularisation, class weighting or sampling, and appropriate treatment of missing values.
- Rigorous evaluation: Temporal validation, rolling backtests, customer/team grouping where appropriate, leakage audits and preprocessing fitted only on training data. Evaluate calibration, precision or lift at operational capacity, and performance across markets and customer segments.
- Model interpretation: Practical use of SHAP and related methods, including their limitations with correlated features. Understand that explaining a prediction does not establish what caused an outcome.

***Advanced expertise depth in at least two of below areas***
- Survival and repeat-event modeling: Kaplan-Meier estimates, Cox proportional hazards and discrete-time hazard models; right-censoring, left-truncated histories, proportional-hazards diagnostics, competing events and horizon-specific calibration.
- Time series and forecasting: Seasonality, trend, autocorrelation, structural breaks and changing exposure periods. Experience with STL decomposition,



forecasting models such as ETS/ARIMA or Prophet, state-space/Kalman methods, prediction intervals and rolling-origin evaluation.
- Hierarchical modeling: Multilevel models and partial pooling for customers nested within sponsors, teams or markets; uncertainty-aware comparisons of small groups. - Hierarchical forecast reconciliation across region, market and global totals.
- Causal inference and experimentation: Randomised experiments, power and minimum detectable effect, confounding, propensity scores, overlap diagnostics, difference-in-differences or doubly robust estimation. Experience assessing incremental impact or uplift where treatment-assignment evidence supports it.
- Customer value and recommendations: Horizon-specific revenue/value models, two-part models for zero-heavy outcomes, item-to-item or basket recommendations, chronological ranking evaluation and cold-start handling.
- Experience with clustering, PCA, robust anomaly detection, drift diagnostics such as PSI, and Monte Carlo simulation for decision uncertainty is useful when applied to a clearly defined business problem.

**What strong candidates should be able to demonstrate**
Show us a project where you discovered a material data or measurement problem, changed the modeling approach because of it, validated the result on later data and explained its practical value to business users. Be prepared to discuss why you chose the target and evaluation design, which alternatives failed, and how the model behaved after deployment.

Pay: ₹110,000.00 - ₹200,000.00 per month

Benefits:

- Flexible schedule
- Work from home

Work Location: Remote

📌 Data Scientist (India)
🏢 Applied Generative Solutions
📍 India