Is this forecast is out of control?

FGS is exception-driven to automate the process of finding and prioritizing the exceptions. If you are forecasting hundreds or thousands of SKUs, you do not have time to review all of your forecasts every period. And if you did, you would likely over-manipulate your forecasts, making them worse.  So you want to review only the most important items. How do you identify them?  With your high workload, you do not want to spend all your time finding the exceptions—you want to spend it analyzing and fixing them!  

FGS uses techniques common to statistical process control (SPC).  Applied to forecasting, the forecasting of each SKU is a process.   Each period the forecasts of a few SKUs go out of control (i.e., they begin to diverge from the actual demand).  One SPC tool FGS uses is the Demand Filter.  It works like an X-Bar chart.  FGS flags the SKUs having a demand beyond a certain number of standard deviations.  Odds are that last period's exceptionally large (or small) demand was not a random occurrence.  You would review the demand transactions and identify the special cause, if any.  You may want to treat it as an outlier or you may want to modify the forecast.  Typically both are done in the FGS Simulation program.

Another SPC process that FGS uses is the Cusum (for cumulative sum of errors) Chart.  Rather than looking just at last period, the forecast error is summed over several periods.  A forecast that is in control should have positive and negative errors that cancel each other out over time.  A biased forecast is consistently over- or under-forecasted.  Over time its error accumulates on the high or low side.

In this example, the model has a slightly declining trend at -4.7% per year.  Over the last few months, the demand exceeded the forecast—not by much, but it has been consistently under-forecasted.  This SKU is flagged as having "hit a tracking signal".

This screen shows the Cusum Chart.  The control limits are set using parabolic masks (blue).  The Cusum Chart looks for large cumulative errors in recent periods and/or small cumulative errors over a longer period of time.  Since the end of 1983, the cumulative error (green) has accumulated on the high side (demand greater than forecast) and has hit the hot limit (red), which is 2.6 standard deviations in this example.

You look at this SKU in the FGS Simulation program and consider a different model.  You increase the pattern change detection sensitivity and you watch FGS detect a change in the pattern of demand.  By ignoring the first year, the model improves, the trend increases, and the error and safety stock decrease.  Since the standard deviation decreases, the lower window's limits are tighter.  So if your theory is wrong, you will hit another tracking signal very soon.  Besides, you note that this is a $1.83 class D SKU and you certainly don't want to under forecast it!  You save the new forecast.  Your (or E/Step Software's) replenishment system may plan a replenishment order, given the increased forecast.