How Artificial Intelligence Is Transforming Warehouse Operations and Supply Chain Management

Right now, artificial intelligence (AI) is probably the most misunderstood top-of-mind technology anywhere in the world. And by all indications, that is not going to change anytime soon.

Many think AI is a black box of universal truth. However, as recent as four years ago, it was probably not as sophisticated as a 3-year-old.

That has changed rapidly, and the technology in general is now about as mature as a typical teenager. Maybe a 20-year-old. Said another way, AI is not a fully formed body of knowledge.

Perhaps it is the rapid rate of change that makes current expectations hyperbolic.

Let’s go all the way back to this summer when the CEO of AI at Microsoft, Mustafa Suleyman, said that in 15 to 20 years AI will reduce to near zero the marginal cost of producing new information. Some took that to mean there will be little to no need for human creativity going forward. Reread this paragraph as many times as needed to let that all sink in.

Just a few weeks later, the leaders in AI, basically the mega-cap tech companies, reported earnings. But more importantly, they reported what they were spending on AI and how those expenditures have already improved the bottom line.

People were generally overwhelmed by expenditures and underwhelmed by return on investment. That’s when this immortal statement was heard: The age of AI is over. Well, that didn’t last long.

Feel whipsawed? You should. Now for some facts.

Where AI stands

Here’s what we know about AI inside the four walls.

It is real (nascent, but real) and will be a major factor in materials handling equipment, systems, software and operations for many years to come.

However, AI right now is highly aspirational in all four. Expect a long rollout of AI in the plant and warehouse.

By example, AI lives in a world right now where its output needs to be checked. Don’t assume AI is always right. Not yet.

At least initially, AI will be a feature as far as people like you are concerned. It may, someday, be individual products like a software package that improves your operations. But let’s not get ahead of ourselves. We’re nowhere near that on any scale.

So, we are in the really early days. While there are sure to be outliers, the typical materials handling supplier company has only worked on AI for two or three years. And only some of those have even got a pilot running at an end user’s site.

Here are some examples discussed by suppliers interviewed for this article.

Dematic expects to work with customers this year on an AI feature to manage load pick buffers for the next few days.

Hyster-Yale Materials Handling is focused on using AI to enhance the efficiency of its lift truck product development process.

Zebra Technologies already includes AI capabilities in its data collection systems at the data analysis stage, looking for patterns people aren’t fast enough to find.

Swisslog has developed enhanced robotic item picking capabilities using AI-powered vision systems.

GreyOrange, which says it has been using AI for at least five years with its autonomous mobile robots (AMRs), now works with other AMR suppliers to teach their robots how to interact in a mixed-supplier facility.

DHL Supply Chain, which centers its AI efforts in its 200-person data analytics department, uses AI to improve stocking levels at its third-party logistics (3PL) sites.

Details on all of these efforts in a bit.

Surely, these companies are just the tip of the iceberg. AI is on its way and going to make things interesting inside the four walls starting yesterday.

What AI delivers

AI in materials handling starts with data collection about your operations. That’s followed by high-speed computational analysis of that data using algorithms. They lead to new observations about what’s happening out on the floor and how to improve. Over time, AI learns as it analyzes more data.

“Quite simply, AI removes some of the cognitive load from people,” says Andre Luecht, Zebra’s global strategy lead for transport, logistics and warehouse. A bit of an understatement, to say the least.

Sounds straightforward, but the transition to AI is no gimme. Just start with the data. AI is sure to change current habits and practices.

According to Deloitte, 97% of data produced by hospitals is not used. Do we really think that profile is much different for DCs, warehouse and manufacturing facilities? Probably not. AI will change that.

Actually, analyzing data results in AI’s output and its value. Often by recognizing patterns in operations unnoticed by people, AI analytics suggest actions to improve those operations.

At DHL, “AI lives under an analytics umbrella,” explains Eric Walters, DHL’s vice president of analytics and performance management on the operations excellence team. You could say his title is a summary of this story.

It’s important to note that DHL is also an outlier (in a good way). It has decided to build its own AI universe and enterprise. That is atypical for you and other end users of materials handling technologies.

Instead, suppliers from Dematic to Zebra will build AI features (perhaps occasionally an AI product) that will improve your operations. At least initially, they will be point-based solutions, not broad solutions.

Now that doesn’t mean you won’t be receiving all the benefits of AI. You will, with a range of benefits.

“AI allows an operation to be more cost effective, precise, visible and efficient,” says Luecht. He continues to tack on to that list: reduced inventory, accelerated operations, increased visibility and optimized use of equipment, systems and people.

Akash Gupta, CEO of GreyOrange, explains the value of AI quite succinctly. “When you can predict better, you can execute better.”

James Sharples, vice president of Swisslog’s global business acceleration, offers some general applications. He says: “Materials handling tasks that can be automated by AI include:

• Relieving humans of monotonous tasks with robotic processes,
• Vision capabilities to replace quality inspections and/or inventory management,
• Planning and forecasting,
• Various supply chain tasks,
• Exception handling,
• Resource management and
• Maintenance.”

AI can be applied to a range of equipment and systems, notes Scott Gaston, managing director and partner at consultant St. Onge Co. These include autonomous mobile robots, lift trucks, robots, automated systems and more. He continues to say that AI will also be embedded in warehouse systems with warehouse execution systems near the top of the list.

“How to leverage AI is a skill set all in itself,” says Gopi Somayuajula, senior vice president of global product development at Hyster-Yale.

He goes on to explain that every answer from AI must be reviewed for accuracy these days. There are no guarantees that answers are right just because that’s what AI spit out, Somayuajula adds. “Just like the AI itself, people have to experiment and learn.”

While those are the specifics, here is a broad framework to put in perspective the major benefit of AI in materials handling. In an interesting way, the potential of AI is similar to that of 3D printing.

One of the most distinguishing capabilities of 3D printing is its ability to make shapes and parts that absolutely cannot be made with any other manufacturing technology. Ever.

Similarly, AI enables operations within the four walls to manage inventory and activities in ways that cannot be done without an AI intervention.

That’s fact. Not conjecture. And you should bet your operational excellence on it. But walk carefully for the foreseeable future.

AI in your facility

Applying AI is going to be a challenge in general, says Gaston.

Equipment and software vendors will enhance their existing offerings with AI capabilities. This is how many facilities will implement AI in their operations, says Gaston.

The question will be which technology to upgrade first and why. Developing an AI implementation roadmap, inclusive of both equipment and systems, will be important so that the end state is purposeful. It may take years to implement each component of the roadmap, but each implementation should move you toward the goal of an AI-enabled operation.

“You have to start with identifying a problem that needs to be solved. Then, you need to match an AI solution to it,” says Gaston.

This part of the decision-making process actually has another leg to it, says Gaston.

“For each application of equipment and systems, a decision will have to be made about the value AI brings it. Will AI offer a sufficient increase in functionality to justify the time, effort and expense of applying AI?” he adds.

Gaston goes on to note that “AI is not a thing in itself. It is a piece of a larger system or piece of equipment.”

In other words, AI for materials handling is not a utility. It is not Microsoft Office. Instead, it is a specific algorithm-driven technology for a specific warehouse or manufacturing application. At least for now.

This is a good point to explain where generative and predictive AI fit here. It gets a little convoluted, but we can do this.

To begin, generative AI is ChatGPT and related products from companies such as Open AI. As Wikipedia explains, generative AI produces “text, images, videos or other data using generative models, often in response to prompts. Generative AI models learn the patterns and structure of their input training data and then generate new data that has similar characteristics.”

Predictive AI, according to Cloudflare, uses “statistical analysis to identify patterns, anticipate behaviors and forecast future events.”

For materials handling applications, predictive AI allows a user to predict demand for a particular SKU based on past demand and other factors. Or, it might evaluate how inventory is arranged in a warehouse to maximize space usage while optimizing accessibility to each SKU based on historical activity.

While generative AI will be associated with materials handling, predictive AI, at least for now, will be the more dominant of the two. Quite simply, predictive AI better suits the skill set needed in materials handling applications.

But don’t be surprised when a particular AI application uses both types. That said, which type of AI is in use is probably of not much concern to most people. But now, it’s nice to know, regardless.

Without doubt, AI is a work in progress.

“The software industry as a whole is not great at building AI projects and commercializing them, but we’re getting better,” says Brett Webster, Dematic’s director of product management.

To back that up, he mentions a Harvard Business Review article from October of last year. It estimated that 80% of AI projects fail.

It is critical to note this number refers to AI projects in general. And it refers to those companies that are building AI products and features, not implementing them.

That said, Webster points out that implementation is not a slam dunk either. In 2022, Forbes said the primary cause of AI failures is a lack of focus on data.

At it a while

Now for the proof that AI is more than a vague concept. Here’s what six companies are working on today—each is in its own lane and timeframe. All six have made major commitments to AI and see it as the future.

Walters says AI was first recognized as important at DHL Supply Chain in 2015 with the addition of autonomous mobile robots to its facilities. Over time, AI gained importance internally and the company added a focus on AI in analytics to its global strategy in June 2023. The company focuses on both generative and predictive AI.

DHL has developed its own generative AI tool, says Walters. “It’s a smart assistant that helps to eliminate open space on blank screens primarily for administrative purposes. I use it every day,” says Walters.

Predictive AI is very active but limited. “We’ve had the resources to develop less than 10% of what we could if time and dollars were unlimited,” says Walters.

That requires heavy prioritizing of projects and benefits matter. “Not the least of which is financial savings. True profit increase as a result of the application is just as valuable as cost avoidance,” says Walters.

Key projects underway are inventory and transportation management tools. At its 3PL locations, Walters says DHL Supply Chain uses AI to better manage e-commerce apparel. “We are using machine learning to improve cycle counting to predict where risk (stockouts) is in the order fulfillment process,” Walters says.

DHL is also making an effort to better orchestrate materials handling. Key are task, resource, inventory and flow optimization, says Walters. “Ultimately, we’d like to present real-time data to operators and shift work as equipment is available,” he adds.

Like DHL, GreyOrange has been at AI longer than most. According to Gupta, AI was actually there when he co-founded the company in 2011. He and his co-founders saw it as a natural progression from their days at university.

Comparing then and now, Gupta says GreyOrange can build an AI application in a week when it used to take two or three months. In addition, the projects are even more ambitious today.

While not connected, GreyOrange’s current major effort in AI is parallel to what DHL is trying to do with its materials handling equipment and systems. However, DHL is focused internally while GreyOrange is working on a much bigger stage.

In short, both are trying to optimize equipment use with a facility. However, DHL works within its own ecosystem. GreyOrange has focused for the past 2.5 years on partnering with a dozen other vendors of autonomous mobile robots. The idea is to have all proprietary systems share data and work with each other using AI. They presented a limited demonstration of the project with seven different AMR suppliers at the Modex show earlier this year. This is one ambitious project. In every way.

And others see this, too. Dematic’s Webster is one of them.

“The vision of delivering autonomous warehouse operations or autonomous supply chains is dependent on the sharing of data,” Webster says. “To achieve this vision, we must find a way as an industry to share operational data such that we can create systems that can be easier to learn and adapt.”

Like DHL and GreyOrange, Zebra has been at AI for a while. But as Luecht explains, rather than developing all AI applications for sensing, analyzing and acting on data in-house, Zebra has partners that provide unique applications for industry problems. “All of these capabilities are enablers and differentiators that become part of Zebra’s portfolio,” he adds.

For the end user, the outcome is the same regardless of the product development cycle, says Luecht.

“AI features in data capture and management of equipment and systems support frontline workers in their decision making,” adds Luecht.

Other AI projects

By now, you should have a sense that AI in materials handling is coming together rather rapidly.

In fact, Swisslog’s Sharples says AI is advancing at three to four times the rate of development than in the past. For his current favorite, robotic item picking, Sharples says AI has advanced in the past six months from being a 5-year-old to a 10-year-old.

In fact, Sharples talks about Swisslog’s use of predictive AI on a robotic item picking project.

The robot, using a vision system, needs to be able to consistently identify different objects mixed in a bin based on color, shape, size and other characteristics.

Predictive AI, through the vision system, then predicts for the robot which object to pick and how to grip and handle it for a clean pick on the first try. In the future, generative AI can be added to provide language-based instructions to the robot.

Sharples continues to say: “This is all about optimizing warehouse resources with AI. A lot of AI effort is going to be focused on facilitating more automation in the DC. That, in turn, drives more sophistication including more sophisticated AI.”

If you liked that, here’s another AI attempt to optimize warehouse resources from Webster at Dematic.

“We are working with a company focused on its SKU demand forecasts. Company growth is high and fixed automation is a large investment already made,” says Webster setting the stage for the AI project.

“The company needs to determine how much more output they can get from what’s already installed. And from that, they need to decide how much more automation they need and when to buy it without hindering growth,” continues Webster. All from AI?

Actually, the key to the kingdom here is building more efficient load pick buffers for today and even several days out. They are in the process of collecting data about SKUs and quantities as well as past demand patterns.

From here, says Webster, Dematic will build a simulation that will lead to AI models for use by the warehouse execution system. The next step is to use the data to build connections between all the moving parts.

Over at Hyster-Yale, Somayajula is focused on the use of AI for lift truck product development. However, he does have a favorite list of future projects.

Somayajula explains product development is hardly new at Hyster-Yale. Yale introduced what is known as the first electric lift truck in 1919, and Hyster was founded in 1929.

However, the majority of the product development process is in engineers’ heads, says Somayajula. If that’s not serious enough, employee turnover and retirements creates a loss of tribal knowledge.

So, the company is in the process of using AI to develop models to enhance the productivity of engineers and improve speed to market.

Hyster-Yale recently sponsored a competition across many universities by the Kellogg School of Management and Segal Design Innovation. Twenty-eight teams worked for six weeks to identify how to reduce development time and effort by 50% using AI.

Going forward, Hyster-Yale has several use cases to leverage its AI efforts. These include: training documentation for maintenance techs, maintenance recommendations for each truck model, and lift truck build and delivery schedules for customers. It’s all about building a corporate AI roadmap and executing it, says Somayajula.

All of that said, Gupta offers three warnings as we run down the AI path.

1. AI requires getting the data right. Garbage in/garbage out.
2. AI is not meant to work in isolation.
3. AI is not an end. It is a means to an end.

To which he added, “humans can only trust their gut so far. Then AI takes over because when you can predict better, you can execute better.”