Any demand planner will tell you some of the most difficult items to plan accurately are those that exhibit an intermittent or irregular demand pattern. This type of demand pattern is characteristic of service parts inventory, wholesale distribution and capital goods. Such items are also referred to as slow moving, lumpy or long-tail items. Statistical methods don’t handle periods with zero demand and tricking these methods by adding a value of one to zero periods simply over-inflates demand. Applying an average demand for an item also has its issues, namely in some periods you won’t have enough and in others you will have way too much. Some intermittent demand pattern items carry a high profit margin so you don’t want to run out. While others can become obsolete quickly so you don’t want to carry too much inventory. In some businesses, intermittent demand items can represent a large percentage of total active SKUs and planning them often consumes significant time and effort.

So what’s the answer? Stochastic Planning of course. Rather than deterministically forecasting demand like most statistical methods, a stochastic strategy focuses on three key issues of the puzzle.

1. Generating a good estimate of the likelihood (probability) of a demand occurrence within the replenishment lead-time.
2. Obtaining a view of lead-time distribution for the service-level inventory requirement.
3. Analyzing the risks of understocking and overstocking to make a good replenishment decision.

Demand occurrence likelihood or probability is determined by evaluating historical demand data of an intermittent item and its frequency pattern of demand occurrences. The appropriate probability formulation should be adaptive to the inter-arrival variation in demand, also known as the coefficient of variation. Demand probability should also consider how long ago the last demand occurrence happened. As with most real data there tends to be a component of noise that needs to be filtered out to reach the true inter-arrival pattern. The use of stochastic simulation models can be very helpful in understanding the intricacies of the coefficient of variation.

In contrast to exponential smoothing methods, a stochastic method provides a full-spectrum view of lead-time demand distribution. To figure out lead-time demand distribution and derive an appropriate inventory policy, one must consider not only customer demand variability, but also lead-time variation of the supplier. A statistical simulation method can be used to estimate the lead-time distribution and to provide guidance to set service-level inventory policies.

With stochastic planning, a replenishment decision is made from the demand occurrence probability estimate, and the lead-time demand distribution. This decision involves whether or not to release a replenishment order and if so the amount to order. Product and business attributes can be used to determine a threshold value reflecting the stock-out tolerance. A replenishment decision is made when the calculated demand occurrence probability exceeds the threshold. An analysis on the risks of both under and overstocking is valuable when choosing the appropriate probability threshold value. For instance, if an item produces a large margin then the threshold for stock-out would be set very high. Alternatively, if an item is expected to go obsolete in the near future the threshold for stock-out could be set very low. An optimization effort can follow to balance risks and minimize overall costs.

Effectively managing intermittent demand continues to be a challenge for supply chain planning teams. A stochastic planning method supported by enabling technology to automate the process can lead to substantial inventory cost reduction, help maintain a higher customer service level, while freeing up significant planner time to work on more strategic and higher value opportunities.

Does your company have a significant number of items that display an intermittent or lumpy demand pattern? Could your planning efforts benefit from implementing stochastic planning?

Additional Reading:

• White Paper: Eight Methods that Improve Forecasting Accuracy
• Blog: Risks in Supply – An Omnipresent Danger