“Moving to AI enabled technologies in cornerstone business processes like forecasting is a high priority in our business transformation agenda. With this project, we’ve come a long way and luckily are now at a stage of making an ML-enabled forecasting business as usual”.
Unilever USA is the biggest and most complex operating company in Unilever managing more than 200 brands. Having such a diverse portfolio which includes beauty and personal care products, food and refreshments, and home care categories, makes sales forecasting one of the most challenging and time-consuming processes in Unilever USA.
Forecasting thousands of SKUs by channels and customers has always been reliant on conventional statistical methods and the expertise of demand planners and sales managers, allowing for human error. This has resulted in fluctuating forecast accuracy and further negative financial implications.
Unilever USA is the biggest and most complex operating company in Unilever managing more than 200 brands. Having such a diverse portfolio which includes beauty and personal care products, food and refreshments, and home care categories, makes sales forecasting one of the most challenging and time-consuming processes in Unilever USA.
Forecasting thousands of SKUs by channels and customers has always been reliant on conventional statistical methods and the expertise of demand planners and sales managers, allowing for human error. This has resulted in fluctuating forecast accuracy and further negative financial implications.
Machine learning enabled forecasting models were developed to transition a business wide statistical forecasting to a new process powered by the most advanced ML algorithms. Within the project, custom models were created to generate forecasts for forecasts for the US and world’s biggest retailers.
For each custom model, there was a unique set of input data including point-of-sale sales, inventory, promo investments, the number of retail stores, and national holidays. The richness of the internal and external input data helped algorithms learn quickly and generate a forecast with higher accuracy and less effort.
The Domo and Anaplan platforms were chosen to help accommodate the new forecasting process end-to-end for top customers. These platforms assist with everything from running the code for all models, disaggregating it and making corrections to visualizing the forecast for all end users. Having the brand-new ML forecast visualized on interactive dashboards allowed demand planners and analysts to understand what influences each forecast driver. This understanding sped up the adoption process significantly.
Apart from custom models, a universal ML baseline forecasting model was developed and launched to cover baseline forecasting processes across all categories and customers. Thanks to the special desegregation dashboard at Anaplan, demand planning managers can verify their newly created ML-baseline by drilling it down to specific SKUs separated by retailers, compare with previous years and much more.
Machine learning enabled forecasting models were developed to transition a business wide statistical forecasting to a new process powered by the most advanced ML algorithms. Within the project, custom models were created to generate forecasts for forecasts for the US and world’s biggest retailers.
For each custom model, there was a unique set of input data including point-of-sale sales, inventory, promo investments, the number of retail stores, and national holidays. The richness of the internal and external input data helped algorithms learn quickly and generate a forecast with higher accuracy and less effort.
The Domo and Anaplan platforms were chosen to help accommodate the new forecasting process end-to-end for top customers. These platforms assist with everything from running the code for all models, disaggregating it and making corrections to visualizing the forecast for all end users. Having the brand-new ML forecast visualized on interactive dashboards allowed demand planners and analysts to understand what influences each forecast driver. This understanding sped up the adoption process significantly.
Apart from custom models, a universal ML baseline forecasting model was developed and launched to cover baseline forecasting processes across all categories and customers. Thanks to the special desegregation dashboard at Anaplan, demand planning managers can verify their newly created ML-baseline by drilling it down to specific SKUs separated by retailers, compare with previous years and much more.
