Consider eliminating outliers

An outlier is a demand which is unusually large or small (relative to the rest of the demand).  If you know a particular outlier could not recur, you can save inventory by eliminating the outlier when computing the model.  You screen for outliers using the outlier sensitivity.  Currently the outlier sensitivity is set to eliminate historical periods with residual error that exceeds 4 standard deviations. You, the analyst, consider ignoring an outlier and change the outlier sensitivity to 3 standard deviations (a tighter limit, thus increasing the likelihood of falling outside it).  

To illustrate eliminating an outlier, hover your mouse over the simulation screen. See how the lower window changes.  You will see the outlier flagged and the forecast and error recalculated.  By eliminating the one outlier, both the error and safety stock reduces, in this case by over $40,000.

Now you have a decision to make.  Your choices are:

• Throw out the outlier, save $40,000 in safety stock but increase the risk of a stockout; or
• Keep the original model, invest the $40,000 in safety stock but protect your customer from risk of a stock out if this happens again.
  

Who else can answer this but you?  Let's say you look back at the demand history for the outlier period and learn that month's spike was caused by a special cause.  It won't happen again.  So you save the second window and allow the safety stock to drop.  You might even decide to invest this savings into safety stock for a number of other SKUs that merit higher service.