Scope
- Demand forecasting for a customer in South America
- Looking for a step-change in improvement in forecast accuracies
- Automate understanding of the changes in the demand patterns and ability to get insights.
- Ability to automate demand review collaborative process with Demand planners, demand managers, sales, and marketing teams
- Improvements in storage cost to serve and inventory turnovers through forecast accuracy
Solution
- Data received for 465 products from 2014 – May 2019
- The forecast was conducted for M-1 and M-3 (for Jun-Aug 2019) at two level of:
- Month / Product / Region / Zone / Channel
- Month / Product
- Forecasts generated with multiple statistical (ARIMA, Holt Winter, etc) and machine learning (CatBoost, XG Boost, Random Forest) techniques
- Analyses:
- Exploratory Data Analysis
- Feature engineering and importance
- Forecasts outputs and error analysis (MAPE)
Results
- M-1 Forecast accuracy is 89% at the product / monthly level
- M-3 Forecast accuracy is 84% at the product / monthly level
- Did not have data on prices and promotions
- Insights will improve significantly with such data
