Applying machine learning to optimize standard gross margin in medtech

This thesis explores how transaction-level machine learning can strengthen margin management in a global medical device company facing pressure on its standard gross margin (SGM). Using several years of invoice-line data, it develops an XGBoost model that predicts SGM% based on commercially intuitive factors such as product characteristics, sales channel, and consignment usage, complemented by simple time variables. The model outperforms a traditional rolling-average benchmark and is used to flag transactions where realized margins appear weak relative to model expectations. These outliers are then linked to a realistic set of pricing and portfolio levers to estimate their potential financial impact. The results show that focusing corrective actions on a relatively small subset of underperforming business can improve overall SGM with only a limited effect on revenue, illustrating how interpretable machine learning can be embedded into existing BI and pricing processes as a practical tool for margin governance.