Five Additions to My Network Design Talk
I have a great (at least in my mind1) introduction talk to network design.
The talk has plenty of case studies and stories. It is fun because it is interactive. And it has enough meat to teach some concepts.
I had some extra time this year, so I decided to enhance it. My friends and colleagues who have heard me give this talk many times will ridicule me for keeping most of the old material (I would argue that the old material is classic and too good to remove).
Here are the five most significant additions to the talk.
One, Gurobi’s Burrito Game.
The game is a fantastic way to teach network design. It shows the difficulty of picking locations as close to as much demand as possible while balancing the fixed cost of adding new burrito trucks.
I use it at the end of my talk. I say, “this is what we just learned about.”
The game adds about 20 minutes to the talk. The game has a competitive mode. I recommend that the students like it.
Two, hidden objectives
This section is really about multi-objective and hierarchical optimization.
Other parts of the talk cover multiple objectives— cost versus service being the big one. I now formalize it and frame it in terms of hidden objectives.
For example, you often hear rules like “all my customers need to be within 100 miles of a warehouse.” This might be better analyzed as a multi-objective problem. If you can get 90% of customers within 100 miles with half as many warehouses, that may be a better business decision.
Or another example is when lines on a network design map cross each other (see map below).
People hate to see lines crossing. But, the lines may cross for many valid reasons. Uncrossing the lines often involves a hidden objective that people only realize when they see the crossing lines.
In the picture above, the stated objective was to minimize cost. But, once the managers saw the crossing lines, they realized that they also wanted customers assigned to the closest warehouse for simplicity. They were willing to allow the costs to increase by up to 1% for an easier-to-implement solution.
This is done with hierarchal optimization. I like hierarchal optimization because it comes closer to replicating how managers make decisions— almost everyone is willing to give up a little bit on a primary objective to gain a lot on a secondary objective.
Three, Jeff Camm’s idea for analyzing risks with scenarios
My network design talk already touched on risk. Jeff Camm of Wake Forest had a great talk on risk at a GE Research conference. It inspired me to beef up my section on risk.
The big idea I used from Jeff was his idea of coming up with many different scenarios and then determining the optimal network configuration for that scenario. Then, you take each optimal configuration and test them with all the other scenarios. What you are looking for are robust solutions across all the scenarios.
What I took from the talk was Jeff’s point that coming up with the scenarios is the hard part. This was an interesting take. The work involves figuring out the right scenarios to ensure you cover the possible risks. I need to learn more about his method, but the idea is good in a network design talk.
Four, the idea from Paul Williams’s book on getting value from building a model
In Paul Williams’s book on Mathematical Modeling, he mentions that you can often get as much value from building an optimization model as you can from getting the results.
This is a great point. Many years ago, at a CSCMP conference, someone from Mars/Wrigley talked about how they found $10M in savings by building the baseline model.
You will learn a lot about your supply chain by building a model of it.
In his book Serious Play, Michael Schrage makes a similar point— building models can tell you a lot about an organization. It is worth paying attention to the model-building process.
Five, Deloitte’s case on locating a fab plant
Deloitte was recently involved in a project to locate a new fab plant for Micron. When you locate a fab plant, things like tax incentives, access to talent and universities, and the surrounding community are much more important than transportation and logistics.
So, this makes a good example of where you don’t need optimization.
I could be biased in liking this talk. But other people say they like the talk. And I get invited back to the same places to give the talk again. And I’ve definitely given lectures and talks that have fallen flat; this talk doesn’t fit in that category.