Opinions differ on inventory: a lean proponent might argue that inventory belongs to the objectionable 'waste' and should therefore be eliminated. However, for many companies, stock is essential - not only to enable efficient production, but also as an essential part of customer service. The question is therefore not so much: to stock or not to stock, but rather: how much stock should there be, of what product, and where?
Systems for smart inventory management
The good news in inventory management is that there are quite a few theoretical models that can be easily applied in practice. Whereas advanced planning systems (APS) largely lean on arriving at a solution interactively with the human planner, systems for smart inventory management typically go a step further. A simple model that assumes normally distributed demand and a fixed delivery time only needs a desired delivery reliability to calculate stock levels.
Standard inventory management models often not applicable
The less good news is that there are many situations in practice where standard models are not applicable anyway. For example, if the demand for the product is not normally distributed, as is often the case in service supply chains. Another assumption that often doesn't hold true is the fixed lead time - because it depends on how busy the supply chain is, or on disruptions. And if you have a supply chain with multiple disconnection points, it is no easy puzzle to determine which disconnection point to put how much stock in. It can be figured out, though, using standard optimisation engines that use linear programming. The mistake made then is that such models cannot deal with uncertainty - they consider the demand to be deterministic. Suppose you have two products with predicted demand, you have limited capacity, and one product yields more than the other. Then a deterministic optimisation will fully prioritise the more profitable product, while practice usually shows that you need some stock of the other one too, if demand for the profitable product lags unforeseenly once.
Deterministic models
At Eindhoven University of Technology, colleague Ton de Kok is working on techniques that can deal with uncertainty - so-called stochastic demand, and there are some nice examples of the application of these techniques. Unfortunately, they are not yet able to take variable leadtimes, or capacity, into account as well. Sometimes that is surmountable, and if capacity is an important factor, there is nothing else to do but to get to work with those deterministic models, and then, via some tricks, still get a robust solution out of them.
Vincent Wiers - Expert Planning & Scheduling
