Summary
The Model Adjustments feature allows you to make future adjustments to the statistical model, while still leaving an item on auto-forecast.
It is accessible in the Forecast Manager by clicking Edit --> Model Adjustments.
Common scenarios for using model adjustments are for when you have advance notice of a large change in demand for an item, to the point where past history will no longer be a good indicator of future demand, meaning that the statistical model would be inaccurate for a long period of time before catching up to the "new normal."
With Model Adjustments, you can put your expected adjustment in place and let that be your statistical model, and when the stat model has enough history to "catch up" to the new level of demand, the model adjustment is automatically retired.
Adding / Editing
To add a model adjustment, open the model adjustment window, and click the "+ New" button. Only one model adjustment may exist for a given hierarchy node at any given time, so if one already exists, you can (re)create it.
Enter the values for your adjustment:
- Name - A name to help you remember what/why you're making this adjustment
- Start Date - when the change will begin, e.g. when is the new demand coming in to play?
- Adjustment Amount - the quantity or percent increase that you are predicting due to this change.
- Adjustment UoM - whether your adjustment amount is in terms of units or percent change.
- Note - any additional notes you wish to leave for yourself.
Optionally, you can provide information about the initial fill-in of the adjustment. For example, if the adjustment is due to a new customer or new store opening, it is reasonable to expect that they have to fill their supply chain, which can lead to an unusual bump in the first few forecasting days or periods. If so, StockIQ can create an event to predict this and also to account for it once this bump has moved into history.
- Fill-In Adjustment - What amount of additional initial fill-in uplift you are expecting
- Fill-In UoM - As above, whether or not this is framed in terms of units or a percent.
- Fill-In Length - what is the duration of time (# of calendar days) over which you expect the unusual fill-in behavior to persist. This will be spread over the appropriate number of forecasting periods.
When you are done, click Save. The adjustment will be immediately calculated and applied.
Adjustment Life Cycle
When you create a model adjustment, typically it is at some point out in the future, say January of the next year.
When January arrives, and StockIQ sees that you are now in the period where the adjustment is first going to take effect, a snapshot is taken of the statistical model with your model adjustment applied. This snapshot of your current model is used as your statistical model for this first 3 months (or 6 weeks for weekly) no matter what, INCLUDING this first period.
Starting in the 4th month (7th week), StockIQ will begin to compare the error of the snapshotted statistical model (with your adjustment applied) versus the current statistical model. If the snapshotted model is more accurate than the current "naive" statistical model, the model adjustment snapshot is used.
This comparison continues until the model adjustment is either more than six months (13 weeks for weekly) old, OR until the current "naive" statistical model is more accurate versus recent demand history (since the start of the adjustment) than the snapshotted model with the model adjustment.
When either of these conditions becomes true, the model adjustment is marked as "Complete" and is in effect retired so that is no longer active and having any effect on your statistical model moving forward. The adjustment will still be visible in Forecast Manager, but will show the "Is Complete" checkbox checked in the user interface.
Model Adjustment vs Manual Forecast
In some ways, the model adjustment has a similar effect to simply making a manual forecast adjustment. with an auto-forecast start date in the future.
However, there are a few advantages:
- For adjustments starting further in the future, your forecast periods until the beginning of the model adjustment continue to update, whereas in a manual forecast, these become fixed-in-time.
- Additionally, the model adjustment continues to adjust on top of your baseline stat model up until the stat model is snapshotted at the beginning of the model adjustment's start period.
- The model adjustment will auto-decide when best to retire itself based on accuracy vs recent actuals, versus auto-activating on a fixed date.
Troubleshooting
Model adjustments will not impact the forecast unless adjustment are made at the level of the forecast hierarchy which is designated as the control in Default Forecast Settings > Default Forecast Level.