One of the first assignments in my modeling career was to re-design the European distribution network for a leading Electronics manufacturer. I was a rookie modeler back then, and the primary goal for me in that project was to learn from the lead modeler.
Apparently, another consulting company had worked on developing a model for the client a couple of years ago and we decided to go through that model before we embarked on our own modeling journey. After going through the Model tables for less than an hour, the lead modeler mentioned to me that he knows the individual who developed that old model for our client. I didn’t get it. My question was-how can you tell that, just by going through a model? His take was, that just by looking at the way some of the constraints were created and data was aggregated, he could guess who it was.
As per him, Analytics professionals, over their careers, develop their own style, just like artists. If you are a seasoned Analytics professional yourself, you can identify the work of another professional, if you have worked with them before.
When he mentioned the term artist, I was confused. Back then, Supply Chain modeling was all about science to me- applying math to business problems. Fast forward to 2016, having spent nearly a decade in supply chain and logistics domain and many years in advanced analytics and consulting, I can now relate to his statement. Supply Chain Modeling, I believe, is both Art and Science.
Modeling practitioners are well conversant with the science behind Supply Chain Network models. Even though many Supply Chain modeling applications these days don’t require you to fully formulate the entire Linear Program, having the capability to visualize the problem on hand in terms of an objective function, decision variables, constraints etc. (i.e. as a Linear or Mixed Integer Programming problem etc.) helps you define and develop the model to get the results you want.
So it may seem that Supply Chain modeling is all about science. However, my perspective is that Supply Chain Network modeling involves a deep aspect of Art as well. You have to be artistic as well when developing those awesome Supply Chain models. Whenever I mention that there is an element of art involved in network modeling, it seems to baffle some of my colleagues.
Based on my modeling experience, I now believe that transforming a business process into a mathematical formulation is not entirely a scientific process because often, it is practically impossible to capture every constraint or factor that can impact a business process that you are trying to replicate in your mathematical model (not to mention the data availability issues that modelers run into frequently). This is where the science part of modeling hits a roadblock.
The Art aspect of Network Modeling is developing Network Models that can provide good enough outputs which can be used for strategic decision making, despite not being able to capture all the constraints or despite not going into analysis paralysis (trying to develop an extremely detailed model, even though that level of detail may not be required). Each modeler, over their career, will devise/develop their own “Art” of making a model mimic the business process, with or without the availability of all data elements and without going into the same granularity in every model.
In the Illustration below, I have tried to capture how both Science and Art elements together help us develop Supply Chain network models that deliver desired outputs and help us make strategic decisions.
I will briefly touch upon only the Art elements depicted in the illustration above as Science aspects are more Black and White whereas Art elements are subject to individual perception/style of each modeler and I want to share my perspectives here.
A. Avoid Analysis Paralysis: My definition of Analysis Paralysis, in Supply Chain Network modeling context is a scenario where you start over analyzing data so much that you get into a data/modeling quagmire. Few ways to avoid analysis paralysis in Network design projects have been briefly touched below:
A.1- Differentiate between important and trivial: It is important to understand what portions of the business processes/Supply Chain you can exclude from the model and yet develop a model that will serve your purpose. A good example is, if a portion of your network will not be impacted, it may be a good idea to exclude that portion, even though that portion is a part of the overall network you are trying to model (for example, if you don’t intend to use the output to optimize Inbound, exclude it from the model).
A.2- Determine the right level of details for your model: Determining the right level of data depth for your model can lead to models that are less complex and solve faster. A great example is data Aggregation. You can aggregate Products, Customers, Time periods and cost types, based on the requirement of the model. You don’t always need to model at SKU level for high level strategic decisions. There are so many ways to get creative with data aggregation, depending on the objective of the model.
B. Link output to Strategy: Your model is essentially a math problem and it will throw out all those fancy numbers at you. Some of those numbers will make sense right away whereas some may seem weird. Don’t blame the model! It is doing what we asked it to do. As a seasoned and well rounded modeler, you need to decipher the output and transform it into something that links to the strategic objective of Network modeling exercise.
B.1-Eliminate the noise in the output: We always have to keep in mind that all the model can think of is to achieve its objective function. If it is a cost minimization problem, the model will go to any length to save every Dollar it can save. In doing that, it may sometimes make weird decisions, say, move 100 cases from one DC to another to save $100 . If you have an extensive and complex distribution network, you will find such noise in your output most of the time. You need your business judgment to eliminate noise elements from the output.
B.2-Combine Qualitative with Quantitative factors: Once you have a clean output, you need to tie these numbers with real world business aspects. For example, the model may be suggesting cost savings of x million dollars for a scenario, but then there may be other factors that can come into play like Labor Unions, Disruption costs, public relations etc. that may not make that particular scenario very attractive.