In a recent episode of the podcast Supply Chain Optimizers, Amaon’s Carlos Armando Zetina from Amazon had a great conversation with Diego Solorzano.
Here are five lessons about supply chain engineering inspired by the episode:
One, not all supply chains need to be “hands-off.” Part 1
Carlos pointed out that most supply chains are unlike Amazon’s and don’t need a hands-off approach. A scalable, hands-off system is an expensive engineering feat. If your corporate strategy doesn’t need it, the return on investment likely won’t be there.
This reminded me of something a data science leader at Netflix said in a talk. He mentioned that everyone wants real-time analytics. Why not? It is cheap to ask for it. However, it is expensive and complex to deliver real-time analytics. A data science leader must ensure that the application needs a real-time response.
The same thing is true about a hands-off supply chain system. It sounds good to say, but it is expensive to deliver.
Two, not all supply chain applications require a “hands-off” approach, Part 2
Supply chain applications can be classified as strategic, tactical, and operational.
A typical strategic application like network design can’t be hands-off. Before you acquire a new building or shut one down, you have to test many different scenarios and discuss the results with many people (executive leaders, finance, legal, etc).
A tactical solution, like setting inventory levels, might have more automation but still require someone to review and override. Carlos made a good point that it is impossible (or very expensive) to gather all the data you would like for this decision. The managers and planners will have data that is not in any system.
By the time you get to operational decisions, like scheduling a factory, you may have even more automation.
The level of automation should be an engineering and business decision.
Three, we think in terms of supply chain solutions, but the engineering to create those solutions requires many algorithms.
This point resonated with me. As Carlos said in the podcast, any supply chain solution has many “moving parts.” This requires many different engineering decisions: what parts do you aggregate and disaggregate, what are the different algorithms needed, and how does it all come together?
But, just as important, how do you keep such a system up-to-date when business strategies change?
This is why companies need strong supply chain engineers and a way to systematically maintain these solutions.
Four, simple is better.
Carlos has an academic background and mentioned that he thinks there is a gap between the complexity needed to publish in academic journals and the engineering task of implementing solutions.
He has a five-minute rule: It isn't simple enough if it takes longer than five minutes to explain the algorithm.
Five, consider all your levers.
This last point is about digging deeper to uncover new levers.
For example, many demand forecasting systems just use historical sales data. But, if you can build forecast models with price as a variable, you’ll have more levers to pull in the supply chain. If inventory is low or capacity tight, a higher price will lead to fewer sales. This may be the right decision for the overall supply chain.
I agree with 4 in theory, but have you tried explaining how MRP works in detail?