In our previous article about selecting a replenishment solution, we explained that forecast quality is the most critical aspect to obtain the expected results. We have prepared an illustrative example to provide a better explanation. First, we will calculate the forecast, we will sell some units and finally, we will calculate the forecast quality.
A company sells 2 products A and B, which are bought from the same supplier. The supplier requires a minimum order of € 1,000 to serve the customer. For simplicity, we will assume that it delivers immediately. Product A costs € 100 and Product B costs € 50. We are in week “0”. We have 5 units of each product in stock and we would like to keep at least 5 units of each of them in stock.
Calculation of the forecast
The program has generated the following sales forecast for the next 4 weeks:
Calculation of the optimal order
The program calculates that to complete the minimum order of € 1,000, the best option is to buy the demand for weeks 1 and 2. Furthermore, by doing so, we would not have to order the supplier again next week.
As the supplier serves the orders immediately, stock gets updated as shown in the next table.
Reality contradicts the calculations of our program and the actual sales for the first week were as follows:
- Sales Product A in W1: 9 units
- Sales Product B in W1: 0 units
After sales, stock at the end of week 1 is:
- Stock Product A at the end of W1: 3 units
- Stock Product B at the end of W1: 11 units
And we realise that we should order product A again because we only have 3 units when we would like to have 5 units. Additionally, we need to cover the sales of week 2. Although you may think that this situation is extreme, it is prevalent for many companies.
Forecast error calculation
The forecast error has been:
The bias is calculated as forecast – sales and is a value expressed in units.
We have calculated the error in percentage as | Sales – Forecast | * 100 / Sales. In the case where sales are 0, we consider that:
- If the forecast is 0, the error is 0%
- If the forecast is <> 0, the error is 100%.
Companies work with indicators and that implies working with aggregate indicators. We have frequently observed that forecast aggregated indicators are obtained in a way similar to the following:
SURPRISE! The aggregated bias is -2 units, but individually, both deviations were higher, in the opposite direction. We find the same behaviour with the error in percentage.
The individual error in each product was greater than or equal to 56%. When calculating an aggregated indicator, it seems that error is only 22%. Thus, it can be inferred that the quality of the forecast is 78% and that the result is fantastic, when it is not.
A common wrong conclusion could be to think that problems are solved by investing 22% in safety stock. In this case, no sales have been lost, so adding safety stock would only amplify the problem of excess stock for item B.
In another article, we will discuss the most important shortcomings of existing formulas to measure forecast quality, especially when they are used in an aggregated way, and we will present our proposal to measure forecast quality with a model better related to the impact on the business and work with more valuable indicators.
Author: Eduardo de Porras
Publication date: 31 May 2021