Aptitive was able to use data science to increase the accuracy of a client’s manufacturing demand forecasting by over 10%

The Problem

An international manufacturer was struggling to make supply chain decisions due to inaccurate manufacturing demand forecasting the variability of their product demand and unknowns in their supply chain. A large majority of their products include electronic components for various industrial applications that are purchased in large quantities by both consumers and distributors. The nature of large order sizes, lack of any recognizable seasonality, and long lead times added to the challenge of forecasting. Prior to this engagement, forecasting was done at an individual SKU level based on an average of the prior, day, week, month, quarter and year sales quantities. A number which did not reflect any other influences or nuances resulting in forecasts that could not be trusted, lengthy manual decision making processes and ultimately lost profits.

Our Solution

Aptitive built advanced manufacturing demand forecasting using machine learning models to forecast demand by week and month for the clients largest and most volatile products. Aptitive introduced a process that utilized statistical packages and machine learning methods in R in order to drive more effective forecasts. We underwent a process of feature selection, model analysis and outlier analysis in order to ultimately develop a set of time series models that decomposed changing trends and volatility of the client’s past sales demand.

The Outcome

Our process and models enabled an 8% increase in accuracy for weekly sales demands and a 12% increase in monthly sales demand accuracy for the largest of the client’s product offerings. The impact of these resulting baseline metrics alone was such that the client has undergone an initiative to implement similar measures across all of its divisions and components of its supply chain.

Client Industry

Electric Manufacturing