I've re-joined the AIChE (American Institute of Chemical Engineers), and one of the benefits is their monthly magazine, Chemical Engineering Progress, which looks a lot different than it did when I was a member five years ago.
In the September issue there is an "Ask the Experts" column on Managing a Responsive Supply Network (pdf) by Ray Adams of SAP. He covers a lot of familiar ground on the impact of variability and predictability of the supply chain. But he doesn't take it to the next step in discussing what one could / should do to mitigate all these factors.
There has been a lot of work in chemical engineering circles (and others) on optimizing the production at individual facilities and across entire networks. As Adams indicates in his article, however, the traditional mode has been to produce whatever can be produced and shove it into the supply chain. This has the guaranteed effect of having a glut of stuff that no one needs AND absences of things that the supply chain wants.
What are the variables in the supply chain? (Adams again): demand forecast, production rate, transportation times (upstream and downstream). In multi-product plants (and supply chains), the optimization problem can get fairly complex fairly quickly. On the customer end, there are also variable time from a perceived need for the product and placing an order.
Why are these things a problem? Typical practice is to attempt to forecast demand out many months -- much longer than the time needed to actually make the products -- so that the plant can have as much flexibility as possible in arranging the "optimal" schedule for equipment utilization or other metrics. When variability rears its head on this kind of time frame, you are guaranteed to have overstocks AND shortages at the same time.
In this case, the problem can be significantly solved by decomposing it. The time for getting a product from the plant to the customer is some version of time customer waits to order + time to manufacture + time to ship. If there is a layered distribution network, there are multiple order-ship cycles in this case. The way to reduce this time (I am assuming a manufacturer wants to reduce the time from raw material to delivery to the customer), is to check each element and see what is happening.
Forecasting is mostly a black art. Use it to drive longer-term marketing efforts, but not the production schedule. But, but, but you need that forecast to decide what to make! No, just make what your customers are buying. This means your production time and your distribution network can move quickly in response to those needs. To do that, the entire supply chain needs to know what is being consumed at the end of the chain, and it should work in concert to pull goods through as they are sold. In other words: remove the customer-order wait time. Make it daily (or instantaneous) with electronic reporting. The transportation time variability is what it is, but shouldn't be a significant portion of the lag. The big work has to do with shifting the manufacturing facility to be more agile and flexible, working with smaller batches, so it can respond more rapidly to demands of the market. I suggest that Theory of Constraints can work in the plant and in the network as well.