Time series forecasting has several applications within manufacturing businesses ranging from helping plant managers to set up and streamline their production schedules to helping executives arrive at meaningful budget estimates.
We have detailed the case of one small manufacturer that has been using sales forecast data from their highest volume products to help setup their production planning operations. By simply using historical sales data one can forecast with reasonable accuracy the demand trends for the next several months. With such a basic model, the manufacturer has been able to predict monthly demand with 80%+ accuracy. More reliable has been the trend prediction using the model, which has been able to capture the up and down swings in the demand with nearly 100% accuracy as seen in the charts below.
We also have the case of a mid sized manufacturer using time series data to improve their budget forecasting. By using a two step process, first predicting an aggregate cost using a regression model, and next forecasting this aggregate cost using time series analysis, the executive team in the logistics group was able to forecast their monthly transportation budgets with 94% accuracy.
There are several choices for a forecasting model, typically a Holt Winters model which has trend and seasonality indices can do a good job. We can also use hybrid methods, such as the ones available within RapidMiner, which allow applying any machine learning algorithm to a time series. Of course, a quick and ready forecasting option is available with some visualization tools such as Tableau.
Extending a production planning forecast models to a more general budget forecasting forecast model involves aggregating the forecasts for all products a company manufactures. Addtionally, this aggregate model can be used to segment analysis by region or customers. Finally, incorporating scenario projections by modeling the effect of macro economic trends on a company's demand will make the tool a very useful for budget forecasting.
Once models like these are built, it takes very little effort to continue updating them and therefore the case for using them regularly in making business decisions becomes stronger. Furthermore, once these models are built and validated, there is very little to no technical skills needed to continue using them.
But there are many challenges a small business might face in utilizing such value adding tools. The first one is certainly the expertise needed to put in place the models to begin with. The second challenge might appear to be the technology needed. However, in the last several years, with the maturation of so many high quality open source software tools, this obstacle has been clearly overcome. The real challenge lies in building the awareness of executives from small and medium businesses to the possibility and potential of such value adding tools. Once decision makers are made aware of the business potential of these tools, the technical challenges are easily overcome.
Download the two case studies mentioned above to get more details on the process and business value.