In the first part of our blog series, we explained the basics of forecasting models in material planning and examined their advantages and disadvantages. In this second part, we focus on consumption-based forecasting, an approach that heavily relies on actual consumption data to determine future material needs. Consumption-based forecasting models are particularly useful in environments where material consumption is relatively constant or follows clearly definable patterns, such as seasonality or trend developments. We will take a detailed look at how this approach works, its advantages and disadvantages, and its potential applications.

Types of Consumption-Based Forecasts

Consumption-based forecasts rely on historical consumption data, analyzing patterns and trends to predict future needs. Since they focus on a single variable—consumption—these models are also referred to as univariate models.
Depending on the type of material, consumption patterns may vary. Consumption does not necessarily need to occur frequently to exhibit regularity or trends.

Consumption-based forecasting uses continuously updated data, extending the time series to improve calculations over time. Depending on the accuracy and quantity of the data, patterns and trends can be identified.

Certain calculation methods are generally better suited to specific patterns. Typical methods and models in the field of consumption-based forecasting include:

  1. Average: The average is the simplest form of forecasting. It calculates the mean of all past data points and uses this as the prediction for the future. This method is particularly useful when the data is stable and shows minimal fluctuations.
  2. Moving Average: The moving average considers only a fixed number of the most recent data points and calculates their mean. This helps smooth out short-term fluctuations and better reveals the underlying trend. For instance, the average of the past six months might be used to forecast the next month.
  3. Weighted Moving Average: In a weighted moving average, more recent data points are assigned higher weights than older ones. This makes the forecasts respond more quickly to changes in the data. For example, the most recent month might be weighted twice as heavily as the previous month.
  4. Linear Regression: Linear regression analyzes the relationship between a dependent variable (e.g., consumption) and one or more independent variables (e.g., time). It calculates a line that best fits the data points and uses this line to predict future values.
  5. Seasonal Linear Regression: Seasonal linear regression incorporates not only linear trends but also seasonal patterns, such as monthly or quarterly fluctuations. This enables the inclusion of seasonal effects, like higher sales volumes during the holiday season, in the forecast.
  6. Simple Exponential Smoothing (Constant Model): This model uses a technique where older data points are given less weight than newer ones. The weighting decreases exponentially, allowing the forecast to respond quickly to current changes.
  7. Linear Exponential Smoothing (Trend Model): An extension of simple exponential smoothing, this model is used to create forecasts when there is a trend in the data. Unlike simple exponential smoothing, which is suitable only for data without a clear trend, this model accounts for both the level (current base value) and the rate of change (trend).
  8. Seasonal Exponential Smoothing (Seasonal Model): This model extends exponential smoothing to include seasonal components. It accounts for not only current trends but also seasonal patterns. For example, it can incorporate typical seasonal fluctuations in sales figures.
  9. Trend-Seasonal Exponential Smoothing (Trend-Season Model): This model combines trend and seasonal pattern analysis. It is particularly useful when data follows both a long-term trend and recurring seasonal patterns, enabling highly precise forecasts.

Choosing the Right Method for the Right Situation

Fortunately, these models do not require manual calculations, as they are machine-supported. However, a basic understanding of mathematics and statistics – such as knowledge of data, variables, and their relationships – is essential. Additionally, one should have domain knowledge, like understanding why and how much is consumed.

Consumption-based forecasts are a powerful tool for predicting future material needs, particularly when historical data patterns exist and can be analyzed effectively. They offer a wide range of models that can be applied depending on the requirements and data quality, allowing for the consideration of both short-term fluctuations and long-term trends. The choice of the right model depends heavily on the specific situation and available data.

In the next part of our blog series, we will focus on implementing forecasting models using SAP. Stay tuned to learn how to make your material planning even more precise and efficient!