Outliers and/or demand spikes are points in an items historical demand that are out of the normal distribution. These are often caused by a one time order event that results in a significant spike in demand.
The challenge with such a demand spike is if left unchanged, it can influence an items forecast and may influence an items safety stock calculation if the standard deviation or root mean squared error methods are employed. Thus, a spike will cause the deviation to increase thereby increasing the safety stock quantity to compensate for the potential of a similar occurrence in the future. As a result, in certain cases, it makes sense to remove these outliers/spikes.
All planned spikes i.e. orders that are planned in advance or special ordered by customers should be removed. These spikes are typically a new product launch and as such are not part of the normal product replenishment cycle. These are typically not pulled directly from inventory.
Spikes for orders that are pulled from current inventory should not be removed particularly for A business importance and high or medium order frequency items.
The article below describes the two primary methods to remove spike: Manual or Automated
Please note that if S&OP is enabled, spike removal in the manner described in the article only applies to safety stock calculations when the Weekly Dev. Safety Stock or Actual to Forecast Safety Stock methods are applied. If S&OP is enabled, the method to remove outliers for demand planning / forecasting purposes, the steps described in the article Editing History in a Demand Plan should be used.
The image below is of a major spike. DemandCaster has two methods to remove spikes/outliers and/or editing of history:
- Manual: The user edits the spike out of the history by clicking on the historical period that contains the spike and then manually editing the spike down to reasonable level. Even though this is a slow process, we like this method because the edits are performed on a case by case basis.
- Automatic: This is a faster process that removes all outliers automatically per the number of outlier iterations selected. The more iterations, the smoother the historical demand and the lower the safety stock. As such, we recommend a maximum of 1 outlier iteration per analysis at one standard deviation. This does not remove all the natural volatility of the product but does remove at least one spike from history thereby moderating safety stock a little.
Clicking on the tip of the spike automatically takes you to the historical period.
Option 1: Manual Spike and History Removal
Click on the 0 next to the identified spike period to open up the pop-up to edit the value
- The pop-up opens to show the history in weekly buckets.
- The user may type in the value they believe the history should next to the period requiring editing. A negative value reduces the original value. A positive value adds to the original value.
- Click save to apply the new historical value.
Option 2: Remove a Customer From History
A second manual option is to remove an entire customer from history. This can be done by un-checking the customer. This will remove all of its demand from history causing the Forecast and Safety Stock (if applicable) to regenerated.
Please note that if S&OP is enabled, the customer / item relationship is not removed from the demand plan. This must be managed in both places presently.
Option 2: Automated Outlier Removal
When the automated outlier removal is enabled in Forecast Settings by entering a value of 1 or above, DemandCaster will automatically calculate the amount of history to remove to address an outlier.
The Outliers Iteration setting specifies how many times the system will attempt to smooth history. The more iterations the smoother the history and the lower the standard deviation becomes. If the use of this feature is desired, we do not recommend more than two iterations. Setting the value to 0 disables the outlier setting. The Sigma Threshold sets the number of standard deviations for DemandCaster to identify an outlier. The higher the settings the more standard deviations will be used.
The method is also described in the article How does the automated outlier and detection work in DemandCaster