Playbook - Demand Planning

Demand Forecasting / Demand Planning

To plan you inventory effectively, you need a forecast of what you will be selling and using of your items.

StockIQ generates a forecast for your item by looking at demand history, and using a statistical forecast algorithm to generate projections into the future.  This algorithm takes into account seasonality, monthly variation, trends, and more, to generate your forecast.  This forecast is generally ~2 years in length.

GOAL: All forecasts are wrong, so our goal is to minimize the forecast error with our efforts.

Your effort in demand planning then will be in one of two main forms, either:

1. Applying human insight to help the statistical algorithm, in the form of Events and Model Adjustments, OR
2. Using Manual forecasts when the historical data is insufficient to accurately represent expected demand.

Configuration Guide

• Forecast Settings - We recommend system defaults of auto-forecasting using the StockIQ Forecasting algorithm, and moving items to manual as necessary
• Under & Over Forecast Report Settings
  • Start with tolerance factor of 4 sigma to avoid getting too many alerts at the start.  Raise to 5 if necessary to avoid being overwhelmed.
  • Set Minimum Unit and Dollar limits as appropriate that is worth your time.
  • Enable alerting on over- and under-forecast, as well as new items
  • Disable alert on low-volume items, at least to start.
• Forecast Vs Statistical Alert Settings (In Manual Forecasts screen)
  • Start with a tolerance factor of 4 sigma to avoid getting too many alerts at the start.  Raise to 5 if necessary.
  • Set Minimum Unit and Dollar limits as appropriate that is worth your time.
  • Enable alerting on over- and under-forecast, as well as new items
  • You may want to disable alerting on Low-Volume, New, and Sporadic items to start with.
• Forecast Error History Alert Settings
  • Set minimum error percentage to 100% to start
  • Check back one quarter
  • Set minimum dollar and unit values as appropriate.
• New Item Alert Settings
  • Enabled

Monthly Process Guide

• Workdays 1-5
  • Alert: Review "Past Forecast Versus Actuals" in the Forecast Vs Actuals Screen - this alerts you to previous months where forecast is off from actual demand.
    • Reevaluate any manual forecasts and possibly revert to statistical/auto
    • Reevaluate choice of forecast algorithm
    • Add human insight from your forecast meetings via events or manual forecast changes as necessary to maximize forecast accuracy
    • Consult with Sales, the customer, or other factors to determine the cause of deviation
  • Alert: Review "Forecast vs Model" Alerts in the Manual Forecasts Screen
    • Adjust poor manual forecasts as necessary to reduce forecast error.
  • Alert: Review "Forecast Error History" Alerts in the Forecast Error History Screen
    • Apply events and/or enter manual forecasts as necessary to reduce error on high-error items based on possible reasons:
    • Limited History? When there is not much historical data available with which to generate a statistical forecast, error will naturally be high.  We recommend using the Forecast Wizard to set a short term (6-12 months) forecast in this case.  
      • If the item is expected to be highly seasonal, you can try setting the seasonality to use the seasonality of the product group it is in until it has enough of its own sales history.
    • Inappropriate Forecast Algorithm Selection? If the selected algorithm is not generating sensible results, you can try selecting a different algorithm
    • Unexplained Randomness in Data? If the sales history shows a large amount of randomness, it is possible that some of the “noise” is unexplained blips.  Promotions, natural disasters, or other confounding factors.  “Explaining away” this behavior using the “Events” feature will help restore the model’s ability to forecast this item accurately.
    • True Randomness in the Data? If the sales history has a lot of randomness that the forecast algorithm cannot explain, error may simply be high because of the nature of the data set itself.  In this case, you may simply have to manually forecast your item or simply accept that there is large variation inherent in the sales of this item, and carry larger safety stock as a result.
    • Item is Sporadic? If the item is just sporadic in nature, then no forecasting algorithm is going to perform well.  A better approach is to set the replenishment order policy to something like Min/Max, Sporadic Sales, or Buy To Order.
• Workdays 5-10
  • Continue with forecast entry and maintenance
  • Alert: New Item Alert
    • Enter forecasts for New item introductions
    • WARNING: 75% percent of excess inventory is acquired on the first purchase order, so strong communication with sales and customer representatives is important!
• Workdays 10-20
  • Alert: Review "Current Forecast Vs Actuals" - this is an advance warning when your current month sales are outside of your forecast based on your operational forecast.
    • Reevaluate your operational forecast if you are manually entering one
    • Reevaluate your choice of forecast algorithm
    • Reevaluate Forecast Model Adjustments and check on adjustments that StockIQ has 'retired' as 'Is Complete' automatically based on comparison of accuracy. For more details see Model Adjustments
    • Try to determine the reason for the unusually large/small sales quantity
    • Is the monthly total going to be too high, or has the sales pattern changed from normal?
    • Investigating the cause of the deviation should let you know if it is expected due to an anomaly in the normal buying patterns, or if an actual change in demand has occurred.
  • Review stock list and ensure that any parts marked to be obsolete are either forecasted to zero, or enter their replacement/supercession using Item Site Replacements

Red Flags

The following are red-flag indicators that all is not well.  For help with any of these, reach out to StockIQ.

• Large number of manual forecasts
• Zero manual forecasts
• No events in the system / nobody knows what events are for
• Distrust in the system-generated forecasts for a majority of items
• No forecast hierarchy configured / missing category information.
• Zero suspended "Past Forecast Vs Actuals" alerts

If you are experiencing any of these, please contact StockIQ and we will help!

Now that we have our demand planning process down, we can start our replenishment planning playbook.