Essentially all inventory stocking decisions can be resolved as a forecasting problem.
This is because the essence of inventory management is determining the most appropriate level of inventory to hold, to service the expected future demand for that inventory, based on the expected supply constraints.
Thus all inventory management requires a forecast of both demand and supply in order to establish the buffer that needs to be held to match these two factors.
This is no different with MRO and spare parts except that demand based on random failure events is, by definition, impossible to forecast. The following is a discussion of techniques for forecasting spare parts and their applicability to MRO and spare parts inventory management.
Two Methods for Forecasting Spare Parts
All forecasting methods can be grouped into one of two classes:
1. Extrapolation of historical data
2. Causal or predictive models
Extrapolation of historical data can vary from simplistic to highly sophisticated but all historical data methods are based on the premise that the future can be predicted by looking at the past. The methods are typically quantitative and can appear to be rigorous but the accuracy is driven by both the validity of the fundamental premise, that is, that the future can be predicted by looking at the past, and the quality of the data rather than the sophistication of the modelling.
Causal or predictive methods can be either quantitative or qualitative. A quantitative approach might rely on forecasts of future planned maintenance activities and the expected usage of parts for each activity. A qualitative approach might rely on the opinions of people involved in the parts usage and procurement. For example, a maintenance team member wants to have a spare part stocked and their view of how many to stock is based solely on their opinion, on the day that the decision is made, of what is ‘safe’. Unfortunately, this happens too often with spare parts inventory and the natural conservatism of maintenance personnel results in significant overstocking.
This aspect of causal or predictive methods often makes people think that all such approaches are unscientific and less accurate than historic data driven approaches. However, this is not the case because causal approaches and the use of forward-looking information is more appropriate for deciding future inventory holdings than relying on the extrapolation of history.
Forecasting Methods Based on Extrapolation of Historical Data
The strengths and weaknesses of the extrapolation of historical data, with respect to spare parts inventory, are most easily explained by reviewing some different methods. It is worth noting that all of these methods manipulate history and present the result as a forecast.
Moving Average
This is the simplest form of forecasting. This approach assumes that a reasonable estimate of future demand can be determined by using an average of the demand for the most recent periods.
Exponential Smoothing
Exponential smoothing is a refinement of the moving average. It attempts to capture the dynamics of changing demand by assuming that the most recent period gives the best insight into demand and each preceding period gives less insight. Essentially it requires a weighting of the demand in each period based on how recently it occurred.
Linear Regression
This approach, sometimes referred to as the ‘line of best fit’ is used when demand exhibits a clear trend in demand (either increase or falling). Linear regression also averages the demand from previous periods but it is poor at handling variable or seasonal demand.
Multi-Factorial Methods
Forecasting algorithms have been developed that attempt to accommodate all of the major demand variation factors (age of data, trends, seasonal factors etc.). These models are generally complex but this apparent sophistication does not necessarily deliver greater accuracy because they cannot overcome the weakness of predicting the unpredictable.
The extrapolation of historical data does have a role in predicting demand for some types of spare parts when demand is stable or changing in a predictable way. But care must be taken in use of these techniques as their convenience can make it then easy to apply incorrectly. In practice, the assumption that the future can be predicted by looking at the past has limited validity and thus these methods have limited practical applicability.
Forecasting Spare Parts Based on Causal or Predictive Methods
Causal or predictive methods can use sophisticated models based on detailed causal relationships but for spare parts management it will more likely involve simple, qualitative approaches based on maintenance plans, condition monitoring feedback, and experience.
Which Method to Use?
The choice of which method to use for forecasting spare parts is determined by the situation and the information the forecaster can access. However, as a general rule of thumb, if causal information is available it should be used in preference to relying on historical data.
For information on our Pro Level membership please visit our Pro Level page.
Posted by: Phillip Slater