Objective
A top retailer wanted to reduce the inventory holding costs by leveraging improved demand forecasts and planning. The goal was to develop a tool to forecast future demand with minimal human interventions that would feed back into the system to help generate purchase orders (with store-wise allocations) based on demand projected and ordering frequency of the vendors.
Our Solution
- Our demand forecasting module leveraged an Ensemble approach with a library of algorithms.
- The MAPE on the training set and on the validation is considered for selecting the specific algorithm for a store and/or SKU.
- Additional considerations factored into the demand forecasts included:
- Effect of Day of the Week
- Effect of Promos/Festive Offers
- Stock-Out bias adjustments
- Seasonality Considerations
- Effect of Week of the Month
- Effect of Outliers/Spikes
- Effect of Known Lean Periods
- Effect of Complements/ Substitutes
- The solution leveraged demand forecasts and historical data to compute vendor lead times and fill rates, leading to re-computation and reassessment of optimum ordering quantity, safety stocks, etc.
The solution was built leveraging tcg mcube, our proprietary analytics platform that can analyse large volumes of data to drive accelerated insights. The built-in data models and data science algorithms drive velocity to value for our customers
Project Impact and Outcomes
- A data science driven approach led to improved accuracy on sales predictions and better inventory rules leading to improved inventory planning.
- The inventory holding cost decreased by over 4% without any compromise in service levels.
