Popular software packages have forecast alert monitors. Each software may term it differently:
1. Manugistics calls them lists
2. APO and Dematra call them as alerts
An overworked demand planner typically plans in the order of 1000+ skus each month. That is where these alerts come in handy. You set a pre-determined alert based on the condition meeting a threshold……. Voila! As soon as the month-end job runs and re-creates the models, the alerts kick in and do their job.
You get a report, perhaps emailed to you at night comprised of all those items that exceeded your threshold. Then your job becomes a tad easier. You walk in to the office with your day planned out thinking to yourself:
“I am going to address those 45 SKUs that show a MAPE above 30% first, then address the next 100 that are above 20%”.
Typically the alerts are set using a comparison of new model values to the in-sample historical values. In essence, they address the model fit issue and alert you to specific situations where the models are insufficient or just plain wrong. Quite possibly, the history may have an anamoly as well.
Now ironically I have seen many planners complain that they are getting too many alerts from the software making them practically not very useful. One planner mentioned that his system was spitting out 9000 alerts for a total of 1000 skus?!
This only underscores the importance of thresholds and correct definitions. If you note this cardinal rule, then you will be well on your way to leveraging the exception capabilities of your software:
1. Automated alert measures can only be set for relative performance meausres – MPE, MAPE, or RMSE relative to Average demand, not for absolute measures such as MAD.
2. The settings are just as good as your understanding of the business – the thresholds have to be reasonable in comparison to historical standards.
Also remember that Alerts can be used for another important monthly activity: comparing the difference between two successive forecast runs. DemandPlanning.Net calls it the Delta ratio. Perhaps you are interested in all items where the forecast delta is 25% or above!
Until next time,