Theory of (Supply Chain) Evolution – into Smart Supply Chains
As you know, my theory is there may never be a fully self driving Supply Chain. But we will end up with Smart Supply Chains, where humans and machines co-exist-and have uniquely defined responsibilities. But we know that we are not getting to that level in a single step. The evolution of Supply Chains will be a stepwise process. What those steps will be will, and should be, unique for every organization in my opinion.
But eventually, we will have Smart Supply Chains, where certain decisions will be taken by Algorithms. The types of decisions will be based on criticality level- more critical ones will need review by humans but in some cases, the algorithms will make the decisions.
Logistics Operations- Decisions Galore
Many of the day today aspects in the world of Logistics Operations are instances where someone needs to make a decision- a decision that eventually translates into a “Go” or “No Go” next step. Note that when I say Logistics operations, I am excluding Logistics planning aspects-which are primarily already driven by optimization algorithms. I though see in many instances, these optimization algorithms feeding into the classification algorithms, that will leverage that data.
Classification algorithms (a brief tailored explanation)
I will explain what classification algorithms in the world of Machine Learning are from a Supply Chain and Logistics perspective so that you don’t have to go through all the jargon that is often thrown (sometimes on purpose) at you when the discussion pivots on Machine Learning.
When you take certain decisions in Logistics operations, you have certain options and you end up choosing one of them. An example can be that your have an outbound load staged at a door that has all the product but one pallet that is Inbound, but still hours away.
As an Operations manager, you need to make a call- Wait till the product arrives or dispatch ? The decision classes, every one hour, are: Dispatch or Do Not Dispatch.
To make that decision, you take into account certain data points- how far away is the Inbound, what products are on the load, for which customers, what are the chances arrival delay- and based on that, you assign an overall “chance” – if the “chances” of the outbound load getting delayed exceed a certain notion in your mind, you take the decision to dispatch, even though the Inbound load has not arrived.
This is what a classification algorithm in Machine Learning does. It calculates the “Probability” and based on that, it classifies the outcome into a class- in this case, Dispatch or Do Not Dispatch. All the data points that youtook into account will need to be fed to the algorithm through automated feeds, so that it can then, based on criteria assigned, suggest and/or make a decision.
Explaining application opportunities with examples
So it is the Logistics Operations part where algorithms need to pitch in to make decisions. In my mind, below are some examples:
An Inbound product is currently in transit and an outbound load, that is waiting for this product, a decision needs to be made on whether it should wait after a certain time.
Based on the real time location of the Inbound load, expected arrival and loading time etc., a classification algorithm like Naive Bayes can assign probablity of arrival delays for the outbound load, based on the departure time (multiple departture time options). And if the logistics managers define a cut-off threshold that is fed to the algorithm- the algorithm can then propose whether the outbound load should wait for Inbound or not.
Will an outbound load get delayed ?
This is a binary Yes or No decision. Logistics regression classification methods can be used to predict the “class”, which are “Delayed” or “Not delayed” ,based on certain attributes like lanes, product, carrier, driver (in case of internal fleet), vehicle type, etc.
Will this product get damaged if packed or stored in a certain way ?
This again is a binary Yes or No decision. Logistics regression classification methods can be used to predict the “class”, which are “Damaged” or “Ok”, based on certain attributes like product, Quantity, Primary packaging, Secondary packaging, Transit time, Transit lane, vehicle type etc.
The list is long, and the opportunities are endless. Every Logistics operations scenario, where someone needs to make a “Yes” or “No” decision, can be assigned to a classification algorithm. The key is to carefully define and categorize processes that should be transitioned to algorithms. These assignments need to be made years before you actually get to the true “Smart Supply Chain” level so that humans in the processes are comfortable trusting algorithms making certain banal decisions in their day to day operating environment.
Views expressed are my own.