Using AI for Demand forecasting & Inventory management
Every business looks to find new ways to increase earnings, reduce risks and improve production efficiency. And the first steps to this is managing inventory and forecasting demand. One of leading manufacturer of residential air conditioning in North America wanted to optimise inventory planning, sales quality forecast at product level for 2000 SKUs across 250 geographically dispersed locations. Stock-outs that adversely impacted fulfilment or excess inventory that resulted in high warehousing and holding costs were a cause of concern. What they needed was to improve demand forecasting and inventory management for individual distribution centres. Findability Sciences used their multi-algorithmic time-series forecasting solution to stabilize the volume and value prediction accuracy at 90% and above
A custom forecasting solution was developed using machine learning and deep learning algorithms like ARIMA, ETS, PROPHET, and Neural Network, which were ensembled to obtain higher accuracy against the traditional time-series forecasting algorithms (exponential smoothing, moving averages, etc.). A dashboard was created with role-based access where the client’s team could view the forecasting outcomes at different granularities.
Accurate prediction of sales volume was the client’s topmost priority, for which not only internal, but external and unstructured sourced of data had to also be analyzed. During the initial phase, Findability Sciences improved the prediction accuracy of the overall volume and value sales, stabilizing the same at over 90%. In the subsequent stages, granular forecasting models were developed to predict the SKU-level, volume and value sales for different regions in the US and Canada
The custom forecasting solution helped improve the forecasting accuracy from 70% to 90%. The client, being dependent on a vast network of suppliers for its US based assembly operations, now equipped with the AI infrastructure to accurately predict sales volume, could avoid stock-outs and transport products to the market more efficiently, which optimized the supply chain and indirectly reduced several costs. The adoption of AI-powered forecasting resulted in substantial improvement in accurate predictions and is on its path of achieving significant Financial ROI .
What started as a pilot for 4 regions and 200 SKUs, scaled within a year into an enterprise AI system for predicting sales of 2000 SKUs across 250 geographically dispersed locations.
Presently, Findability Sciences executes over 1 million predictions each month as part of this project – MAPE (Mean Absolute percentage Error) being the chief metric for evaluating the model effectiveness. The forecasting models from Findability Sciences reduced the MAPE from 30% to 10%