Now the big question is how and where to set the buffers and how big these ought to be. In my quest to optimize buffer levels I like to apply various mental models so I can analyze the problem from various viewpoints (if you follow discussions around the subject on the internet you will find out quickly that the topic is quite complex and supply chain dynamics are not necessarily easy to understand or intuitive). First of all I have come to the conclusion that chasing forecast accuracy is not very benefiting. Even If you’re lucky enough to work out the ideal plan (with very high forecast accuracy), what is it good for if your suppliers can’t deliver to it? It does seem nonsensical to shift priorities from trying to increase forecast accuracy to worrying about your suppliers, but don’t you think that when your supply comes in regular and reliable that you have a lot more control over your inventories and fill rates?
So if the intention is to level the demand to our production schedulers and/or outside suppliers then we must absorb as much noise as we can. Some suggest to de-couple parts from demand using buffers. That works in some cases but when you need visibility into future changes, this can fatally mess with your plan. I personally believe in absorbers, cushions, whatever you want to call them. In a cushion noise is absorbed to a degree and if there's a big change coming at you, you'll be notified by your planning alerts. Depending on the situation (high or slow mover, consistency in consumption, value of part etc.) you need to decide on the size of your absorber.
Other than most believe, absorbers can be implemented not only on the supply side but also on the demand side as the following graphic shows: