Predictive analytics was performed to prescribe recommendations.
- Key parameters were chosen from an exhaustive set of attributes such as OG data for existing stores, Point of Sales data, competitor information, market factors, and behavioural segments
- Machine Learning techniques like GLM, Random Forest, and SVM were used to predict OG orders for new stores
- Bootstrapping technique was implemented for model robustness
- The algorithms were tested and validated recursively on 100 random samples
- The model predictions improved over time
- The solution helped identify factors inﬂuencing OG orders such as the client’s grocery share in CMA, percentage of shoppers who fall under primary grocery households, grocery sales over the past 4 months, OG awareness in CBSA, etc.
- Based on model predictions, the client was able to classify stores as super-high, high, and medium, allowing optimal budget allocation for rolling out OG in select stores