Improving Portfolio recommendations using Conversion

This project focused on enhancing portfolio recommendations for retail stores by leveraging advanced statistical and machine learning techniques. Using a Total Distribution approach powered by CUPED (Covariate-Adjusted Means), I developed a backend solution to measure the impact of portfolio recommendations over a 4-month period.

The project also included designing a conversion attribution model for the South Africa region using GLMNET and Statsmodels Logistic Regression, achieving an accuracy of 98%. Additionally, I implemented a conversion package to calculate incremental lift with a model error margin of ±5%.

By delivering actionable business insights and clear stakeholder presentations, this project significantly improved decision-making for portfolio recommendations, driving measurable growth and efficiency in retail operations.