Consumption Intelligence Engine Using Clustering And Time Series Modelling

November 7, 2020

American multinational food, snack and beverage corporation

Problem Statement

• Lack of visibility into future consumption resulting in sub-optimal inventory management, often leading to overstocks/ understocks
• Low volume of flavor shot add-ons – lost dollar opportunity

Solution Approach

• Data Input
• Model Dataset
• Dynamic K-means clustering to identify store clusters with similar consumption
• Time series forecasting for predicting consumption
• Item-Item Collaborative filtering based recommendations

Business Impact

• Forecast Accuracy across millions of transactions
• Increase in flavor shot add-ons, generating incremental revenue

Critical Success Factors

• Identification of key product opportunities
• Inventory Management