sw-data-scientist_skill

Overview

This skill provides a practical toolkit for statistical modeling and business analytics using Python. It focuses on A/B testing, causal inference, customer analytics (CLV, churn), and time series forecasting to support data-driven decisions. The skill is an archived collection of versions that emphasizes reproducible workflows and clear experiment design.

How this skill works

The skill inspects raw data, runs exploratory analyses, and guides phase-by-phase processing from feature engineering to model training and evaluation. It includes standard statistical tests for experiments, causal identification strategies, survival and cohort methods for churn, CLV calculations, and forecasting models for temporal data. Each response is structured to produce one work phase at a time to keep large analyses manageable.

When to use it

• Designing and analyzing A/B tests or randomized experiments.
• Estimating causal effects where confounding needs careful handling.
• Calculating customer lifetime value and churn risk for retention strategies.
• Generating short- and medium-term forecasts for revenue, demand, or engagement.
• Creating reproducible, modular analytics workflows for stakeholder reporting.

Best practices

• Follow the chunking rule: produce one analysis phase per response for EDA → Feature Engineering → Modeling → Evaluation → Recommendations.
• Pre-register hypotheses and define metrics and sample size before testing.
• Check and document assumptions for statistical tests and causal models (overlap, unconfoundedness, stationarity).
• Use holdout sets or time-aware cross-validation for forecasting and churn models.
• Report effect sizes, confidence intervals, and business-relevant metrics, not just p-values.

Example use cases

• Running an A/B test analysis to measure lift in conversion and estimate required sample size.
• Applying propensity score or difference-in-differences methods to measure campaign impact.
• Building a CLV model to segment customers and prioritize retention spend.
• Forecasting monthly active users or revenue for the next 6–12 months to inform budgeting.
• Estimating churn hazard rates and creating targeted win-back strategies.