Traditional single variable forecasting algorithms look just for basic trends not causes. They utilize patterns like seasonality or trend patterns to provide more accurate forecasts. These algorithms can be very good at picking up patterns and doing predictions for steady runners. However, if there are multiple factors impacting the forecast, as is the case normally, then ML can help to significantly improve forecasts and even explain the results as well as causality.
We have found that just adding a small amount of explanatory data and ML logic to determine which data to use and how (regression and categorization logic), can increase forecast accuracy by at least 10 percentage points. It can be even higher for products that are new, or heavily influenced by external factors. ML is not going to have a big impact on a high-volume steady seller that has very predictable seasonality, but for high value volatile products, new products, and mid to lower volume products, the impact can be significant, especially for more accurate short term forecasts. Some examples of factors other than shipping data that can be used by the ML algorithms are point of sale data, product lifecycle stage, price, holidays, population statistics and industry trend data.
When it comes to supply chain planning strategies, a typical application for ML is generating more accurate demand forecasts as described above. However, ML can benefit supply side planning just as much as demand planning, if not more. Examples are managing product portfolio, deciding service levels for products to optimize profit, determination of safety stock levels, prediction of accurate processing times based on attributes or products and equipment for more accurate planning, or even predict when supplies will arrive based on supplier behavior. For more information of application of ML in supply chain planning and how it can improve operations click Here.
Want to learn more contact us at firstname.lastname@example.org