Understanding your Business’ Variability
Knowing the Difference Between Natural Seasonal Variability and Forecast Variability
By Ryan Rickard, Director SCMO2
Understanding your business’ variability matters! To effectively manage your supply chain and know your forecast potential, it is important to understand your demand variability. Analyzing your historical demand and historical forecast performance are two ways to measure your forecastability and the variability of your forecast error.
Some businesses have more stable or consistent historical demand—the demand, orders, or shipments from customers. Some businesses have seasonal demand—peaks and valleys in their yearly historical demand from customers. Are the peaks in June and December because of regular seasonal patterns in your customer demands, or promotional activity spikes that occurred because of a price drop?
By studying your historical demand, you gain an indication of how easy (or how hard) it is to forecast your business. Also known as forecastability. How effectively do you forecast your business given your normal customer demand patterns? If your natural variability is 25% for the year, the absolute best you could forecast your business is 75%. Forecastability is the inverse of your variability. Yes, some products or customers have differing patterns, so it is important to measure the variability consistently at a level that aligns with your other supply chain metrics and goals. Otherwise your data will tell you two different things.
Measure Variability with CoV
So how do you measure variability? Using Coefficient of Variation across a period of data points (time) will calculate variability. Also referred to as CoV, Coefficient of Variation is defined as the ratio of the Standard Deviation (σ) to the Mean (μ). Measuring the dispersion of the data points around the mean helps show how much volatility (or uncertainty) there is in your historical demand. As mentioned above, we often use the term “Forecastability” when referring to CoV.
Inputs to CoV:
Standard Deviation (σ) of Sales Order History
- shows how much variation or dispersion there is from the mean
- a low Standard Deviation means the data points are close to the mean; a high σ means the data is spread out over many data points
Mean (µ) of the Sales Order History
- the average of all your data points
- calculated by summing up the values of each data point divided by the number of values
Reviewing a CoV Example
In the above example, normal variability across the 12 periods is 19.51%, setting the forecastability at 80.49%. The best possible forecast performance for this dataset, using history as the driver, is about 80%. What does this tell us? An 80% forecastability is good! It reinforces using our history as our future forecast input. This dataset has relatively low variability and is statistically easy to forecast.
But what if your business has regular or normal seasonality? What if your business has routine peaks and valleys because of normal market patterns? An example would be a business that sells products purchased more of during the Back to School or Christmas season. These seasonal patterns naturally create variability across time.
Calculating CoV with High Seasonal Demand
If you have seasonal demand, then a different variability metric called CoV2 or CoV of Forecast Error could be useful. By analyzing how well you forecast for that seasonal variability, we can measure the dispersion of the Forecast Errors compared to the average to highlight how much volatility there is in your Forecast Errors. CoV2 is defined as the Standard Deviation of the Forecast errors compared to the mean of the Sales Order History.
Inputs to CoV2:
Standard Deviation (σ) of the Forecast Error
- shows how much variation or dispersion there is from the mean
- a low Standard Deviation means the data points are close to the mean; a high σ means the data is spread out over many data points
Mean (µ) of the Sales Order History
- the average of all your data points
- calculated by summing up the values of each data point divided by the number of values
Reviewing a CoV2 Example
What are the benefits of studying your CoV or CoV2? You gain an understanding of historical variability, which helps you know how accurate your forecast can possibly be. It allows you to properly set targets or expectations regarding your possible forecast performance. You can group products to focus planning attention. You can segment them into XYZ categories based on variability results and attack the X’s one way and the Z’s another. The advantages are plentiful.
By focusing on your CoV2 results, you learn where to commit resources for improving those combinations with high forecast error. Perhaps you are using the wrong statistical forecast, or maybe the input received is not valuable or was added at the wrong level of detail.
Calculating CoV and CoV2
You can calculate either quite easily in Excel, or by using SAP IBP. For a more detailed demonstration of how to set up these calculations, you can watch our past webinar titled “SESSION I: Variability Matters – Calculating Demand Variability Both Manually and Within the SAP IBP Platform.” This 50-minute in-depth session guides you through all the aspects of calculating a valuable CoV or CoV2 result.
Remember though, as you begin to analyze your variability, we encourage you to start with CoV. As you mature in your analysis you may then want to also study CoV2 to see where your forecast error or performance could be improved. Begin simple and master the segmentation opportunities first using historical CoV before reviewing CoV2 to further improve forecast performance.
To Recap:
- CoV simply measures the dispersion of the historical data points around the mean. It helps show how much volatility (or uncertainty) your historical demand has.
- Standard CoV does not consider forecast error, therefore forecast timing misses are not compounded across multiple periods (like they could be with CoV2)
- CoV2 compares the final forecast to the actuals to analyze the dispersion of the forecast error to the historical demand.
- CoV2 is better for seasonal businesses “if” forecast error is low.
- If you miss the timing of a promotion or seasonal forecast (by a couple weeks), you could generate a forecast error in multiple months.
- As you get better at forecasting, the errors will go down as will the CoV2.
- This does not mean that the variability of the history is better, but you are better are forecasting that variability.
Congratulations! Your forecasting will improve with this added technique. Just a small increase in forecasting accuracy can make a huge impact on cashflow and profitability, so be sure to use all the tools available in your efforts.
