• Tags:: 🗞️Articles
  • Author:: Stefan de Kok
  • Link:: Should You Forecast Monthly or Weekly? | LinkedIn
  • Finished date:: 2021-09-22
  • Source date:: 2016-11-09
  • Highlights::
    • For long processes, you usually look to quarterly or yearly. For short ones, daily.
    • For short term processes, techniques such as demand sensing (detect short trends quickly) and DDMRP (Demand driven Material Requierements Planning)
      • Where DDMRP What Is DDMRP? | Demand-Driven MRP | SAP Insights is a technique for supply management
        • While MRP is a “push” technique that pushes inventory into the system based on the forecasted need, DDMRP operates differently.

          DDMRP takes variability out of the equation by using “pull” for materials in a demand-driven approach. Rather than relying on forecast accuracy – and buffering for fluctuations in demand and supply – DDMRP tracks actual usage and manages replenishment through a simple visual system. Buffer inventory is only used to ensure the availability of key items that are deemed to be of strategic importance. With the use of DDMRP, there is less inventory overall and fewer shortages.

    • Forecast error comes from mainly from splitting manually a forecast:
      • benchmarks performed for pilots and proofs-of-concept show that by far the largest contributor to forecast error on weekly basis is the splitting mechanism applied to a monthly forecast. Typical additional error due solely to splitting is in the range of 30% to 45% points. When the splitting combines regional-to-location or family-to-SKU splits the error becomes much graver. A common splitting logic we encounter in SAP APO implementations from a monthly/region/SKU granularity to a weekly/ship-from/SKU will deteriorate a handsome 80% forecast accuracy at the aggregate level to a mere 20% accuracy at the more detailed level.

    • So he suggests to move from point forecasts to confidence intervals and perform and optimization regarding the distribution of the prediction:
      • The secret to this increased forecast accuracy is to replace traditional statistical forecasts with more sophisticated stochastic forecasts. A statistical forecast attempts to predict the exact amount of demand for a given item in a given month for a region or shipping location. Compare this to a stochastic forecast, which determines the probability of a particular amount of demand ordered from a particular customer due on a particular day. The how and why of stochastic forecasting is explained in “The Future is Uncertain”

      • Probabilistic Forecasting and Inventory Optimization | Towards Data Science