Podcast April 12, 2024

Will Natural Language Processing (NLP) Transform Warehouse Operations?


In the past few years, artificial intelligence (AI) has promised to reshape industries worldwide, offering novel opportunities for innovation and efficiency. Within the realm of supply chain management, some people argue that AI-powered technologies like Natural Language Processing (NLP) are driving significant evolutions from traditional approaches to warehouse operations and inventory management. Through all the noise, it’s essential for decision makers to understand not just the potential benefits of AI and NLP, but also the challenges that accompany these advancements.

The Benefits

  • Enhanced Operational Insights: NLP enables operators to extract valuable insights from unstructured data sources such as emails, customer feedback, and even verbal communications from frontline warehouse personnel. By converting this language into structured data, businesses gain deeper visibility into inventory dynamics and consumer trends for more effective warehouse operations.
  • Real-Time Decisions: The insights gathered from previously unstructured sources can be leveraged in real-time, enabling decision-makers to make agile and proactive decisions based on the most up-to-date information from their warehouse and customers.
  • Automation: With NLP technology, operators can automate traditionally manual tasks such as data entry and analysis which helps to streamline supply chain processes and reduce operational costs. This helps businesses to focus resources on strategic initiatives and value-added activities.

The Challenges

  • Data Quality Challenges: NLP algorithms rely heavily on the quality of input data. Inaccuracies or inconsistencies in data sources can lead to errors in analysis and decision-making, undermining the effectiveness of NLP applications.
  • Complexity and Scalability: Implementing NLP solutions in large-scale supply chain operations can be complex and resource-intensive. Businesses may face challenges related to data integration, system compatibility, and scalability, particularly when dealing with diverse data sources and formats.
  • Ethical and Privacy Concerns: NLP raises ethical considerations related to data privacy and security. As businesses collect and analyze vast amounts of textual data, they must ensure compliance with data protection regulations and ethical guidelines to safeguard consumer privacy.


Implementing NLP into your warehouse operations could potentially provide huge benefits for a business, but operators need to be aware of key considerations regarding data, scalability, and privacy before making the jump. To learn more about this exciting topic, check out the accompanying podcast episode with LIDD founders Charles Fallon and David Beaudet discussing the emerging topic. If you have questions about how you can prepare your operations for the implementation AI and NLP, reach out directly to [email protected].

🔗 Check out the full episode:

Listen: Anywhere you get your podcasts.

Watch the full video below:

[00:00:06.280] Hey, Charles.

[00:00:06.910] Hi, David.

[00:00:07.560] How are you? Good.

[00:00:08.340] Yourself?

[00:00:08.810] Good, good. It’s the end of the week.

[00:00:10.220] It’s the end of the week and the start of Bixie season.

[00:00:13.810] Well, now Bixie is all year long.

[00:00:16.010] I know, but the start of normal springtime Bixie season.

[00:00:19.050] Yeah. Yeah.

[00:00:19.310] So my hair is all out of whack because at lunch I had my bike helmet on.

[00:00:23.090] Okay.

[00:00:23.530] Did my first big ride.

[00:00:25.440] Not to come or go back home.

[00:00:27.210] No. After I eat, I went to Gillesville Neuve circuit where they’re already getting the grand Prix set up.

[00:00:35.760] Well, I don’t care about the Grand Prix, but it’s fun that you can now bike on it already.

[00:00:39.120] Love it.

[00:00:39.660] Yeah. I’m gonna bike tonight.

[00:00:41.250] And the water levels are high, so the rapids are strong. But that’s not why we’re here.

[00:00:47.520] No, no, it’s the end of the week. It is. Okay. I don’t know if I said that already. No.

[00:00:52.640] Behind the headlines. So what are we going to talk about?

[00:00:56.580] We’re talking about an article from supply chain brain called NLP for natural language processing transforms inventory management.

[00:01:07.950] Excellent. We’re going to talk about AI and in particular natural language processing. I think it’s pretty exciting. And the reason I think it’s exciting is as much as you and I have spent the last 24 months decrying the more vaporware aspects of AI, at some point we have to acknowledge that just because salespeople call everything AI doesn’t mean that there is no AI. I mean, there is some exciting opportunities and it’s time maybe to start talking about some of them with people so that we can give our own perspectives and other people can hopefully join in the comments of the podcast and give some of their ideas. So let’s get into it.

[00:01:54.880] Yeah, and this specific article talks about inventory management, and I find that at least some of the things I read regarding AI and supply chain typically revolves around the inventory management. So we could almost say that topic has been covered. I personally find that the article may not have a lot of meat around how it would do so, but, and I think you share the same example, I don’t know if you want to talk about this or kind of jump into, because one thing we want to do today as well is to, well, talk about other areas or other ideas where specifically the natural language processing could play a role in supply chain.

[00:02:40.910] Well, I think the first place we should begin is always, we always have to remember, you know, it’s like, as I do with everything in life, I’m talking to myself. You know, picture this old man who played with computers as a kid. But now all of this technology has gone to a level of sophistication that I haven’t been able to keep up with. And it might be good to explain to people what natural language processing is. And because when once, once the average person, that vp of operations who’s driving in the car and listens to us religiously, as we know, huge, huge audience, that once they understand the underlying mechanisms, then they can start, you know, their own creative brainstorming around how they can make this useful in their day to day operations.

[00:03:39.910] Fair point. So the natural language processing, the way I would describe it and needs also to be supported, well, it’s categorized as a branch of AI, or at least in today’s world, is the ability to take on disorganized or unstructured, sorry, unstructured data that can come from many sources.

[00:04:02.850] I mean, mostly language, actually.

[00:04:05.230] Yeah, but language could be spoken or written, written in emails, written in, and be able to, well, structure it or at least analyze it and pull out or information from that enormous amount of data that is difficult otherwise for typical computer systems.

[00:04:26.590] Can I go a step further? Of course, because I think it’s really important. If you picture the warehouse manager of today, even if they’ve never directly worked with voice technology, they all know it, right? That is a ubiquitous technology in the warehouse. So we picture now we have the person, the picker, or whatever, let’s just use a picker with a headset on, and through that they have maybe ten to 100 commands. And then the ability to read in digits like alphanumeric characters into their headsets and interface with a computer system with those commands, plus the alphanumeric characters. And I think what it’s important for people to understand and anyone who’s actually done the process of, of setting up a voice system for themselves, what you’re doing is you’re actually, you’re not, there’s no technology processing language. It is simply a sound pattern recognition. So you template your voice. You know, when I, when I see a d, if I see right lid, and it would say, okay, I’m going to ask you to say delta, delta, delta, over and again 20, 30, 40 times until the system recognizes how you say Delta. And then when you say it, that sound gets recognized as the sound for.

[00:05:58.890] Delta, for a specific user, you as a user, for you as a user.

[00:06:02.770] So that you could actually, like, whistle. You know, you say d, I say d, d whistle 25 times. And suddenly that whistle becomes what the computer recognizes as you saying d. Okay. What natural language processing does is it is actually taking the language, the words and the structure of the language, and it’s processing it. And what it’s doing is it takes any word, all the words, and it converts it into a vector, a series of numbers. Think of like a multi road, single columned table to represent each word. A matrix, if you will, if you’re thinking linear algebra. So each word is converted into a series of numbers that can convey not only the literal meaning, but the context. I shouldn’t say the literal meaning of the word, but it allows also for the creating relationships between other words and therefore the context.

[00:07:11.610] Correct. So the NLP or the natural language processing is the ability to capture and structure that information. It does need to rely on whether machine learning or other elements of AI that can then, as you said, create context to understand what a human is asking for.

[00:07:31.810] Right. And the key to that is that ultimately all this language is converted into new numbers. It’s coded into a set of numbers. And that allows us. And the reason I think that’s important is when we start thinking about what the future can be with natural language processing, that’s going to play a very important part, because then it allows you to. Well, we’ll talk about that in a second. So the actual other interesting bit of trivia I think that people would like is while it’s actually natural language processing is something that is almost 80 years old in its inception, like in the first early days of it, where it blossomed, is on our phones and needing translation, which we all know is both imperfect but incredible at the same time. Anyway, so let’s move on.

[00:08:27.160] Okay, well, just because I think what you’re saying with this explanation is that it allows us now to work in. If you talk about the voice picking system, something that is constrained, right. That you have to operate in a very set of hard rules.

[00:08:44.160] And it matching sound pattern to sound pattern.

[00:08:46.870] Exactly.

[00:08:47.550] There’s no language, there’s no meaning, there’s no context, nothing.

[00:08:51.110] And now that this, there’s the potential. Not just the potential, there’s the existence of kind of breaking that down and speaking the way you would to. Like we are.

[00:09:01.090] Yes.

[00:09:01.770] Okay, so current problems that we see that can be solved through this. So you and I had discussions, and I think that you had a great idea to the point. So great that I said, well, do you really want to talk about this publicly? Because it’s so great.

[00:09:22.390] All right, now you’re setting me up. That’s obviously a ridiculous joke, but, yeah.

[00:09:28.570] Let’S I think just jump into this because ultimately we were talking about, well, in the context of distribution activities or operations within the four walls where we right now operate with either voice system or scanning system, and that is being processed by a warehouse management system in order to organize and prioritize tasks. Well, there are limitations. There are things that we cannot do or that are becoming possible if we have the possibility to speak, and that what we’re asking for or dictating can be interpreted and I’ll say matched with existing information and data of warehouse management system.

[00:10:09.970] Yep. Well, I mean, when we think about it like we, there are lots of ways in which AI generally not just natural language processes, but AI generally can or will certainly play an important role in how supply chains operate. The obvious ones, you mentioned this at the start of the show, the obvious ones are we’re going to do better demand forecasting, we’re going to do better just supply chain management generally at that 50,000 foot view. And I would predict that most of the ERP companies and other companies that have large suites of, of software devoted to end to end supply chain management are working on those things. They’re maybe not quite best beta versions of what could be today, but they’re certainly on the road. And then we know this. In the actual material handling system world we have, the autonomous mobile robot only exists because of, because of AI enablement. But when we think about other ways, maybe that we can help improve operations, I think the natural language processing little tiny corner of the AI world is interesting to me because you’re actually much older than people might think. We’ve been auditing warehouses for a very long time, and one of the things we both do is we’ve learned not to believe everybody.

[00:11:57.590] And what will happen is, I’ll ask a supervisor, how often does this happen? First your eyes tell you one story and you think you’re already reaching conclusions about how an operation performs based on what you’re seeing. Then you’re looking for either confirmation or modulation from a supervisor. They’ll give you an answer you already know. It could very well be true in seven times out of ten that it’s a spot on answer. But it’s as much as three times out of ten. The supervisor doesn’t really have a sense of the math of their operations. And their answer is going to be, whatever happened yesterday is what they’re going to report to you as being normal. They understand, or they don’t really understand actually how to give you an average. Right. Right. And then you’ll go on and you’ll talk with the folks on the floor doing the job, the pickers, the fork truck drivers, all the people who make this operation work, and they have an enormous amount of opinion about what’s going right and what’s going wrong with an operation. Well, imagine now if at the moment a transaction is being done, they had the ability to have a, well, Nikola coined this, a suggestion box, right?

[00:13:22.980] So I get to a slot, I find the product in this slot and the product in the next door slot look so similar that in the speed, in my desire to work efficiently, it’s very easy to make a mistake. That’s a classic pick line issue. Right. And wouldn’t it be great if at the, when I, you know, when I’m picking my order, when I get to the slot, let’s say I had a voice, a voice, a headset on, I could just say, hey, this, hey, this item, it’s, it’s, it’s a mistake waiting to happen.

[00:13:57.650] Yeah.

[00:13:58.150] Right, right. And you go through that, and as the weeks progress, this suggestion box of observations from the workers is being collected now before natural language processing, it would be a nightmare, right? Yeah. You could record it, but it would be basically voice memos.

[00:14:16.410] Or you could ask operators, say, at the end of your shift, here’s a pen and paper, please fill this. What have you noticed?

[00:14:23.410] Someone will read it, they’re not going to do anything. But if you made it soon, and.

[00:14:27.510] Processing it would be a nightmare too.

[00:14:29.410] So they’d be annoyed to do it and you’d be annoyed to review it. Right. And so now what we have is we have the ability to take all these suggestions and because the system codes it into numbers, numbers that can be decoded into meaningful text as well, but could be, there’s so many ways you can process that data once it’s captured and transformed into that state, that you could see a couple of things. You could see, hey, I don’t believe everyone’s opinion, but when I get an overwhelming number of opinions, I could have a dashboard that says, hey, these are the sore areas. You have a human intervene and go resolve these problems, because twelve people told me that this slot has got the wrong item in it. You could go a step further, you could empower your WMs or some AI assistant bot that is sitting outside of the WMS. You could empower it to actually make changes without you even knowing that you.

[00:15:43.720] Exist to make a change.

[00:15:44.920] Yeah, I mean, you can, you’d have to set up guardrails and, but you could enable it to at a certain level of problem. If you can validate that problem through some quantitative analysis, you could say just send the instructions via the WMS to make the changes and let’s go. Yeah.

[00:16:06.240] So both on the capturing the information on the spot and being able to process it all the way to interacting with the system, and it’s not to replace it. It really is an enhancement of being able to save a little bit of time here, add a little bit of precision there in the normal functions of a warehouse. So in your opinion, who works on this? Do we think that the warehouse management companies can integrate such a thing? Because you talked about erps and them talking at the 50,000ft view of integrating this in their software. But is this something, you know, how do we get there?

[00:16:54.190] Well, you know, it’s a really interesting question. I don’t know. You know, we know that some of the larger players in the warehouse technology world are really focused on other issues. I think a lot of people walk around and think that the warehouse technology is a solved problem.

[00:17:14.770] Okay.

[00:17:15.400] That’s how I feel. I don’t know, but we don’t really know secretly where the research and development money is being spent in those in the larger houses. I would doubt SAP, which has a very nice warehouse management system, I doubt they’re investing their AI thinking into the warehouse management system.

[00:17:35.450] Yeah, yeah.

[00:17:36.490] I wouldn’t if I were them.

[00:17:37.730] Yeah.

[00:17:37.970] There are other priorities. If I’m Manhattan, would I be doing it? Maybe. But I would still probably be allocating resources to the supply chain suite of software that they have. I don’t know. But what I can say is, just like people predict our need for developers or the efficiency of any one developer is going to increase 2030, 4100 percent, 200% with the help of an assistant. I think warehouse like operation supervision has the same opportunity in turning a supervisor into a super supervisor. And I don’t know who’s going to spend the money on that. But if any venture capitalist is listening, we’re always open to plan this out. We’ve got an army of talented people, people who are thinking about these problems like we are and probably thinking better about them.

[00:18:42.840] Well, that sounds good.

[00:18:44.440] All right.

[00:18:45.080] Anything else you want to add?

[00:18:47.240] No, I want to get on a bike and ride home.

[00:18:50.350] Yeah, me too.

[00:18:51.050] All right, good.

[00:18:51.520] All right. Thank you.

[00:18:52.430] Have a good weekend.

[00:18:53.040] You too.

Let’s build world-class infrastructure together.

Book a Consultation

Are you ready for logistics automation?

Take our readiness quiz to find out!

Begin Assessment