Dr. Watson Says: ...the optimization considers the probabilities of the different scenarios directly and comes up with a solution that is best for all the scenarios at the same time ... What Do You Say? Click Here to Send Us Your Comments Click Here to See Reader Feedback

In graduate school, when my friends from back home asked about the classes I was taking, the class called “Stochastic Processes” drew the most mocking as a sign of the geeky-ness of my studies. 

Although it was an interesting class, I had to agree with them—the class title did make it sound very esoteric.

So, I had only ever used the word “stochastic” (which just refers to randomness or probability) in a few classes and as a bit of running joke.

Then, to my surprise, in the early 2000’s several software vendors started using the term “stochastic optimization” to market their supply chain products.  I’m guessing this was done to make the solution sound sophisticated.  But, now, I see the term being used much more--- it is not mainstream, but a supply chain manager (and not just a grad student) should know what stochastic optimization means.

It is most commonly used in to describe inventory optimization.  When you optimize inventory, you need to include variability directly in your calculations—you need to include demand variability and supply variability.  So, in this sense, when you optimize inventory with any type of decent tool, you are doing “stochastic optimization.”  This is an example of a new term being used to describe something that people were already doing.

In another, more generic use of the term, stochastic optimization can be used to describe a problem where you assign probabilities to scenarios and optimize all the scenarios at once.  That is, the optimization considers the probabilities of the different scenarios directly and comes up with a solution that is best for all the scenarios at the same time.

You can imagine that this type of optimization also applies to supply chain network design.  Instead of running multiple scenarios, you can assign probabilities to each of the scenarios and run everything at once.  For example, you may have a baseline scenario of 50%, a high demand scenario with a probability of 30%, a high transportation cost scenario with a probability of 20%.  When you run this as one stochastic optimization problem, you get a solution that is robust across all three cases.

Usually, when people first hear about this solution, they wonder why you don’t always just run your network model as stochastic optimization problems.  In practice, it can be difficult to interpret the results and explain why the model behaved as it did, and it can be difficult to determine the right probabilities to use for the scenarios. 

Final Thoughts:

With the rise in the field of analytics (more on this in future columns), I’ve seen more people working on bringing more advanced stochastic optimization solutions to the market.  Some of these new solutions may have a big impact on the supply chain.  And, I’m now glad I took that Stochastic Processes class.