AI Insights From FORTNA Chief Scientist

1)  Dr. Meller, some readers might wonder what a chief scientist does in an organization like FORTNA. Can you explain?

Sure, the purpose of having a chief scientist in an organization like FORTNA is to have an overarching view of the application of science throughout the company. That includes several things: our algorithms, data science, simulation and emulation. It could also be, the “older” science of ergonomics and human factors and how that plays into operational performance.

The role of a chief scientist is to have a purview over as many areas of science as possible and bring that information, research and data into the company; this most recently includes examining how AI technology is affecting the company and the industry.

2)  There has been a lot of discussion around artificial intelligence (AI) in the marketplace; in your view, what impact can AI have on distribution operations today?

One impactful thing we can do within FORTNA is to have lots of information available when we answer any question, whether it is in the design phase, the support phase, or the software creation phase. We have so much information we can bring to bear, but it’s in so many different systems. We need a way to look at the data intelligently, and we need a link into all those systems: JIRA, Microsoft Teams sites, emails and everything else.

For example, consider the challenge of customer support. When a customer site has an issue the team on the floor needs a quick answer. The customer support person has a wealth of information that can help to provide the answer, but if the information is not readily accessible, the answer won’t be quick. One of our efforts underway right now is to address this challenge. Technically, we’re taking a large language model and adding some small language capabilities to it by training it with information that we’ve created as part of the design and implementation process. The goal is for someone who has customer support questions to have the ability to get the answers they need at their fingertips through a natural language questioning system.

For example, they might have questions about latency issues with a pick-to-light process. They might wonder:

• What latency should they expect?
• What system controls the pick-to-light process?
• How can they perform a root-cause-analysis for this process?

It’s very powerful to bring all our information without going to multiple sources. I think that’s one significant place we can use it.

3) How would you define AI and its use in the supply chain industry?

At a high level, you can find multiple definitions for artificial intelligence, but in my view, AI is when a computer or robot does something you used to attribute to a human doing.

My favorite example of early AI is MapQuest. Before MapQuest, if you were going on a road trip, you had to get out a physical map and route yourself through trial-and-error, or for those of us who are older, go to AAA, and they did it for you! The point is that this was something a human did, but once MapQuest came out, it was like, aha, a computer can do something a human used to do. To me, that’s why MapQuest was first generation AI.

That’s why there is this evolving line regarding artificial intelligence as AI gets more and more capable – it continues to take things humans used to do and now does it for them with a computer or robot.

When it comes to FORTNA and the supply chain industry, we’re incorporating artificial intelligence into our Warehouse Execution System (WES) because we’re taking on jobs that humans used to do or still do today and using a computer or robot to do them. For distribution organizations to be competitive, they need computers, specifically algorithms, to make faster decisions based on their data. They also need robots to take over onerous tasks that humans shouldn’t do due to safety concerns or no longer want to do.

In the case of our WES, we’re talking about a computer algorithm that is incredibly sophisticated in how it looks at the operation’s state and the available data to orchestrate the release of work over multiple areas of the building. Doing this effectively requires a highly complex algorithm that can take real-time data on the state of the operation and interpret the goals of the operation to produce a plan and allocate the work to execute that plan. It’s important to point out that this is not generative AI in the sense that it doesn’t have a probability matrix behind it to drive it. It’s a very sophisticated algorithm that is applied to something a human used to do—and now computers do it. So that falls under the bailiwick of artificial intelligence.

4) Can you give us an example of how algorithms work in an automated solution like an AutoStore?

Here’s a simple example. At one of the larger retailers we are working with, we’re doing a batch pick out of an AutoStore, which means when we ask for a bin, we want a certain number of items out of that bin to help fulfill multiple orders. Well, in this case, the grid is very large, and so, ideally, I want to bring the bin to the closest workstation, but I also don’t want to ignore the other stations. We need to balance the product flow in real-time to both minimize work for the robots, but also ensure bins are delivered in a timely fashion (and not starve workers at workstations). Our software and algorithms enhance the capability of an AutoStore by choosing a close port while balancing workload across the operation for optimal product flow.

5)  Simulation and emulation are new approaches to testing and measuring a supply chain system’s potential and performance. What is the difference between them, and how can each benefit an organization?

Discrete-event simulation is best used before an Detailplanung is finalized to answer questions about the design and the operation’s performance. For instance, do I have enough buffer (i.e., accumulation) between these two operations to handle the variation in workloads? A good example that we see a lot: Is there enough accumulation in receiving so that I don’t impact picking or other processes when I’m backed up? Or what is the maximum utilization we can expect at goods-to-person pick stations given the variation in the delivery of totes from a shuttle system?

Design questions are best answered with simulation. We like to simulate designs for customers to give them a good baseline on what the system can and cannot do. For example, we can simulate the operation and show that if they increase the rate by up to ten percent, the system will perform optimally, but if you go above ten percent, you might run into some problems downstream. Simulation models abstract or approximate the logic that will be embedded in the software.

Emulation, on the other hand, is best used when you’re trying to test how the actual system’s software will perform for a given operation. When you emulate a system, the software thinks it’s talking to an actual conveyor, when in reality, it’s talking to a digital conveyor instead of a physical one. You run the system in an emulated world in real-time, where one second equals one second (which is just one of the reasons why emulation is not the right tool to evaluate design questions). Emulation models have physics engines and so you can see jams and alignment issues that would result in the field if the control software were not adjusted.

Emulation allows you to examine the system’s operation and performance in detail because you’re using actual warehouse software to run the system. This can alleviate the need to wait until equipment is installed to test software.

6) Innovation and technology are developing so rapidly. How do you prioritize what you and your team are developing and working on?

When we’re looking at new technologies that are in and entering the marketplace it can be a challenge to prioritize. On the design side, we’re at a significant point because there’s been so much investment in our field in bringing technologies to bear. It’s to the point that if I can imagine it exists, it probably does.

I think there are close to fifty different autonomen mobilen Roboter (AMR) vendors out there now, and it used to be only two or three. What we’re fighting now is that it seems there is a new company offering something similar in nature every week with a slight difference in use, design or price.

We have to evaluate what is unique about them because it takes time and effort to create a relationship, and the integration tools and software that are needed to pair with their technologies. There are so many great solutions out there now, and we have a lot of capabilities at our fingertips. Our challenge is pushing the market and evaluating the best technological fits for our customers.

7)  What are some of the mistakes you’ve seen when companies are evaluating new technologies or automation? Is there one common misstep that an organization can avoid?

I would say being realistic about payback periods. It seems that every technology salesperson says their equipment or automation has an eighteen-month return on investment (ROI). Then, we get into the design phase and find that it has an eight-year ROI. How is that?

Both statements can actually be true! The eighteen-month payback can have embedded assumptions around shifts (3), design to average ratio of 1.0 and no need to buy now to accommodate growth in the future. That is, just moving from a three-shift operation to a one-shift operation can move a payback period from 18 months to 4.5 years. Almost all of our designs have some design to average ratio that’s greater than one. Let’s say it’s 2.0. This means you utilize only 50% of the technology on a daily basis outside of peak, but you have to pay for the whole thing. You only get 50% of the labor savings, but you have to pay for 100% of the equipment (in this example). And if this technology requires investment today to accommodate growth later, then you can see where the payback period can exceed even 8 years.

The biggest thing is being realistic and understanding that what you’ve been told about this robot or that piece of automation might not necessarily apply to your situation. This is why you want to work with an organization like FORTNA, which has experts and team members who have designed and have experience with these technologies and can apply a true ROI to your situation and help you to maximize it.

8) What do you see on the horizon for automation and supply chains? Is there anything that is getting you excited?

I don’t know if my crystal ball is fine-tuned, but I will say that robotics are becoming more commonplace and that innovation with these technologies has already started. I think we are at a point where physical work in the distribution centers is becoming less and less desirable. And it’s not just because it’s physically challenging; people don’t feel they’re fulfilling much of a mission by working in one. It’s just a job for a lot of people, and the industry as a whole is having trouble attracting and keeping people.

In the future, I see us designing operations around the idea that we want robots to do these undesirable tasks. And that doesn’t mean more and more complicated robotics. We can use technologies to perform sub-operations in a distribution warehouse instead of designing an operation for one robot to complete.

For instance, at a goods-to-person workstation, a worker does two or three actions: grab or place a tote or box, pick and place an item, and then push the box or tote onto a conveyor when the order is complete. With a robotic design, we can split that into three separate sub-operations. So, a single robot doesn’t have to do all three things; it can just do one and be very good at it.

For this to become a reality, the cost of robotics will have to continue to come down, and it will have to be the right fit for a customer, but as the price of labor rises, organizations will need to take a hard look at what their operations look like now and into the future and we will be designing assuming robotic operators.