A 4-step process to demand forecasting
Updated: Apr 11, 2022
In Supply chain, we say Demand is often a highly volatile parameter and thus it makes demand forecasting both an art and a science.
Demand forecasting forms an essential component of the supply chain process. It’s the driver for almost all supply chain related decisions. While demand forecasting is undeniably important, it’s also one of the most difficult aspects of supply chain planning.
Demand forecasting is important to the supply chain because it is the process by which the strategic and operational strategies are devised. We can think of it as the underlying hypothesis for strategic business activities and the starting point for most supply chain processes, like:
Raw material planning
Purchasing
Inbound logistics
Cash flow, and
Manufacturing planning
Demand forecasting also facilitates critical business activities, like financial planning & risk assessment. Most importantly, forecast accuracy enables retailers to avoid stock outs and over stocking, improve production lead times, minimize costs, increase operational efficiencies, and improve overall customer experience especially when we talk about build to stock and build to order business models.
Common mistakes in Demand forecasting:
Relying on hunches and guesses.
Ignoring historical demand/sales pattern
Avoiding data cleansing
Not using tracking signals i.e. Bias monitoring
Avoiding short term forecast accuracy
A guide for starters!
As a matter of fact, there are different approaches being used by companies across the world for their demand forecasting process, however, we will be looking at the basics of what needs to be done and in what sequence to get the maximum benefit.
1. Segmentation:
The very first step in demand forecasting must be segmentation of our products in which we need to define and build segments and group our products based on their importance and forecast volatility; starting from products of low importance and which are easy to forecast.
We end up with the segment that contains products of high importance and those which are hardest to forecast. Most important benefit this process brings in, is that we can customize/optimize processes and analytics for each segment separately, which helps demand planners to focus on critical parts.
2. Data cleansing:
To ensure you are using the best set of historical data for predicting your future demand, it’s very critical to get the dataset clean which might include:
Outlier correction
Removing promotional impacts
Substituting blank values
Others
3. Determining pattern in the historical data:
This step involves analysing the historical dataset using time series analysis to identify the data pattern. Based on analysis different patterns could be
characterised as:
Level
Trend
Seasonal
Sporadic
We can use various statistical methods to identify these patterns like:
Free hand method
Semi average method
Moving average method
Least square method (Regression)
Based on the identified pattern, next step is to select the most optimum statistical modelling for predicting the future demand.
4. Statistical Modelling:
Once we have the pattern, we choose various statistical models for creating our statistical forecast. Selection of these models depends primarily on the pattern identified in the previous step.
For Level pattern, we can use any of the below statistical models:
Simple average
Moving average
Weighted average
Weighted moving average
Single exponential smoothing
Similarly, for a “Trend” we can use
Double exponential smoothing
Auto ARIMA
For “Seasonal” pattern we primarily use:
Triple exponential smoothing
Auto S-ARIMA
These processes will help us to calculate optimum statistical forecast based on historical demand/sales. We must remember that forecasting is an estimate for the future which can’t be 100% accurate, so as a demand planner our aim should be to minimize the gap between the forecast and actual demand.
High errors in forecast might led either to high inventory at the end of forecasting period or a stock out situation causing low service levels. So, overestimation and underestimation both are not desired.
Forecast error is represented by Et
Et = Dt - Ft
Et = Error during period t
Dt = Demand during period t
Ft = Forecast during period t
We can use various statistical methods to calculate forecast error
Average error method
Mean absolute deviation (MAD)
Mean square error (MSE)
Mean absolute percentage error (MAPE)
Along with these methods it is highly advisable to use “Tracking signals” which help us to identify forecast bias. Biasing helps us to understand if a forecast is being consistently underestimated or overestimated or its free from any bias (which is desirable).
We generally use forecast error as our measuring criteria to select the “Best Fit” statistical model.
Impacts of Demand Forecasting on Supply Chain Management
Improved supplier relations and purchasing terms
Better capacity utilization and allocation of resources
Optimization of inventory levels
Improved distribution planning and logistics
Increase in customer service levels
Better product lifecycle management
3 main roles of Demand Forecasting in supply chain management
Critical in strategic planning of business
It helps to initiate all push processes in SCM
It controls all pull process in SCM
Conclusion
To effectively increase profits and mitigate unnecessary costs, every company need to improve demand forecasting and optimize their supply chain. Today, we have various demand forecasting software that uses advanced analytics and distribution metrics to inform business decisions, improve inventory flow and give the resources needed to bring the businesses to the next level.