I recently wrote about why we’re betting big on machine learning — especially as a tool to improve the decision-making process in supply chain management (SCM). In the wake of that post, I’ve realized there’s quite a lot of confusion in business circles about how ML fits into today’s data toolkit.
So let me try to explain. In this post, I’ll explore the difference between automation, optimization and machine learning. First, let’s start with defining each of the three terms — then use two different use cases to help illustrate how each translates into practice.
What is automation?
Simply put, automation is the process of taking a series of tasks performed by a person or group of people and replicating those tasks through computer logic. The tasks are usually simple in nature and follow some basic logic. For example, simple conditional statements such as “if this, then do that” are great candidates for automation. Also, automation is great when there are defined actions that are performed repeatedly and frequently.
The end state of automation is always known before the process takes place, which makes it different than optimization and machine learning. To automate a process, you know all of the possible conditions and they can be described by a set of rules, or heuristics. Those conditions and rules don’t change. Humans can, of course, perform these tasks but they may be laborious — which is why it’s worth automating.
What is optimization?
The term “optimization” has several different meanings. For this post, I’m referring to using math to identify the best solution to business problem.
While automation allows computer logic to replicate human process, applying optimization will help find the quantifiably best outcome. By applying an optimization algorithm to a complex set of constraints (or rules) and their associated trade-offs, businesses can streamline a wide range of business processes, particularly in the supply chain. For example, a consumer packaged goods (CPG) company might have a large combination of variables from different suppliers (cost, quality, speed) they need to consider when planning the production of their products. Optimization can help supply chain professionals make hundreds or thousands of micro decisions that together offer the most efficient (or profitable) way of manufacturing those products.
Optimization is powerful in that it can consider many real-time constraints and trade-offs in an instant, while it could take individuals or teams hundreds or even thousands of hours to factor in all the relevant data. And that combination of constraints and trade-offs can be highly volatile — so by the time human teams consider all the relevant data, it may be out of date.
What is machine learning?
Applied machine learning is optimization taken to the next level. Instead of looking at only current constraints and trade-offs, a machine learning-powered solution factors in a wide variety of historical and current data and predicts changes to those constraints for a future point in time.
Another way machine learning algorithms can work differently than optimization algorithms is the way they use data. Optimization requires a fixed set of constraints and trade-offs. Machine learning algorithms can consider a wide range of loosely related data — identifying historical relationships between attributes and outcomes, going beyond fixed constraints and into probabilities. However, because the relationships in the data are often loose, machine learning algorithms need training on a lot of data to identify those relationships.
Now that we’ve defined the terms, let’s look at what these three data tools look like in practice.
Consumer use case: The digital map
Back when I started driving, I used to have a stack of regional paper maps stashed into my glove box just in case I got lost (though I almost never used them). Then Google Maps came along, as did old-school car GPS devices such as Garmin. And it forever changed navigation. The evolution of the digital map is a good metaphor for what automation, optimization and machine learning look like in practice. (Caveat: Navigation apps and real-time traffic have leveraged an evolving mix of data technologies over the years. So this example is somewhat simplified.)
Old-school GPS devices, without any real-time traffic, represented automation in action. You could punch in an address, and based on your location, a GPS device would produce a recommended route you should take. Regardless of the day/time or traffic, these disconnected systems would give the same automated recommendation in every instance.
The next level, optimization, came once real-time traffic was added. Software (such as Google Maps and Waze) could optimize your route based on your location and traffic conditions at the time you started your journey. It could also dynamically adjust its recommended route based on changes to conditions. It would continually optimize your route based on current conditions.
Applied machine learning goes a step further. It can predict, based on a wide range of historical data and how future conditions may change, your optimal route based on probabilities. For example, consider what happens in the wake of an accident on the freeway. Five minutes after the accident, a limited backup may register as a two-minute delay. But a machine learning algorithm could understand that it’s highly likely for the delay to get longer, especially for a car approaching the scene from 20 miles away, and also factoring in that traffic will likely get heavier because typical rush hour is about to begin. So, navigation using optimization might only factor in a two-minute delay on your route. But a machine learning solution might suggest it will be a 20-minute delay by the time you get there — and would be more likely to suggest an alternative route.
Business use case: the shipping warehouse
As I mentioned in my last post, we’re particularly excited about the potential impact of machine learning in supply chain management. So, let’s take the shipping warehouse as an example to illustrate the difference between automation, optimization and machine learning.
Historically, humans would take orders, navigate warehouses to find and pull items that were ordered, and then prepare them to be shipped. Automation of this process involves prescribing (to a robot) a list of items to be picked and shipped. By applying optimization to this problem, an algorithm would look at the location of those items and suggest a specific path to most efficiently pull those items for shipping. Optimization might even tell the robot to pick two additional orders that happen to be along the path.
A machine learning solution would not only optimize the path but know that while a robot (or team or robots) is picking a set of items, customers are looking at nearby items on the e-commerce website. And based on probabilities, those robots would go ahead and pick those items as well. When you’re a company like Amazon, shipping huge volume of best-selling products, applying machine learning to the warehouse management can lead to an increase in shipping efficiency and speed.
How can automation, optimization and machine learning be applied to your business?
Automation, optimization and machine learning all have important roles in the supply chain of today and tomorrow. If you’d like to understand how they might fit into your company’s supply chain, let’s talk — email us to set up a free consultation today!
