Traditional demand planning techniques were developed decades ago when distribution channels were few and history was the best predictor of the future demand. Such techniques are inadequate for near-term forecasting in today’s more complex environment, which presents a number of obstacles:
One of the intelligent applications in E2open’s Demand Sensing suite, E2open Demand Sensing examines multiple sources of real-time data. Using machine learning, the application combines these inputs to predict near-term demand with a 30 to 40% improvement in forecast accuracy compared to traditional time-series approaches. Field-proven with more than a decade of use in over 180 countries for $250 billion in annual sales volume, the application is a core part of the demand-driven digital transformation strategies of leading global companies.
Precise Daily Forecasts for Execution Time Horizons
While demand planning typically generates a monthly or weekly forecast, near-term activities such as replenishment are usually planned daily. The traditional approach to bridging this gap is to use simple rule-of-thumb proration logic to convert monthly or weekly forecasts into the daily granularity required by distribution requirements planning. However, this conversion step yields a crude estimate of daily demand and fails to consider other key factors such as the multitude of available signals or how orders in one period affect orders in subsequent periods. With E2open Demand Sensing, organizations receive accurate daily forecasts for the next 6 to 13 weeks. The result is better replenishment decisions — each and every time.
Real-Time Analysis of Multiple Demand Signals
Near-term forecasting must take into account the latest data on open orders, shipments and consumption. Data might also include customer and channel inventory, weather, social sentiment and other demand signals. This information can help companies detect shifts in demand that will affect customer orders. Point-of-sale (POS) and channel data can come from a company’s existing systems or E2open’s Demand Signal Management and Channel Data Management applications.
Machine Learning Pattern-Recognition Algorithms
To make sense of multiple complex demand signals, companies require advanced machine learning pattern recognition algorithms that can determine what is predictive and what is not. Influence factors for each signal vary by product, location and time horizon. The relationships among these factors may change over time, requiring continuous self-tuning.
Predicting new product orders presents a particular challenge for demand planning because of the lack of historic data to serve as the basis for forecasting. E2open Demand Sensing uses clustering algorithms to identify groups of related products so new product behavior can be modeled on that of products that are similar. Machine learning algorithms identify such clusters with more predictive accuracy than is possible with planner intuition alone.
Projection of Supply Requirements
The ultimate goal of near-term forecasting is to create the most accurate picture of projected inventory requirements. After E2open Demand Sensing creates a demand forecast, it can then forecast replenishment based on how daily demand will erode existing inventory. The replenishment forecast can in turn be used to generate supply requirements.