Mirra Bhagawat

Forecasting Key KPIs for Annual Target Setting

• Developed forecasts for MAU, Exports, and MAE using FbProphet, analyzing 3 years of historical data to capture trends and seasonality with a WMAPE of ~7%.
• Curated a heuristic approach incorporating lifts from planned teacher trainings, deployments, and product improvements using pre-post analysis, enabling realistic target setting across countries and segments (K12, HED, D2S) bringing the WMAPE to 10%
• Strategized plan for emerging market India by analyzing the effectiveness of various training types across different school groups to identify the most impactful training methods for specific audience segments.

User Segmentation for Personalized Recommendations

• Collaborated on building a model to classify new users as Solo or Non-Solo based on their first 7 days of activity, including clicks, exports, project types, and interactions with collaborate CTAs like Share and Publish.
• Enabled personalized recommendations to improve user engagement and platform experience.

Demand Forecasting for an Indian Liquor Brand

• Developed a range of Machine Learning models encompassing algorithms from Linear Regression to Time Series methods like SARIMAX, FbProphet and Moving Averages, effectively forecasting secondary sales for 1600+ SKUs with a WMAPE of ~20%
• Designed a methodology to forecast sales for new and low-sales SKUs by leveraging the client's Annual Business Plan, enhancing accuracy and improved supply planning.

Promo Effectiveness for a Cosmetics Company - USD 1.2Mn to 2.5Mn SOTP

• Developed a linear regression model to differentiate baseline and incremental quantities attributed to promotions, utilizing factors like paycheck cycles, special events etc.
• Calculated uplift and conducted ROI analysis to gauge promotional impact, leading to informed recommendations for optimal discount percentages and event durations.

Credit Risk for a leading Indian Bank - Uplift of 10X

• Identified segments of entities to group and model based on their default rate and population size using clustering on product attributes
• Developed Application and Behavior Scorecard models to predict Probability of Default for SME entities using logistic regression for different segments using features ranging from Assets data, Liability data and Bureau data with performance metrics Gini, KS, Lift