What Execution AI actually does in the Warehouse

Speaker A: Welcome to Logistics Business Conversations, the podcast exploring the ideas and insights shaping today's logistics industry. Featuring leading voices from across the sector, we examine key trends, emerging challenges, and the innovations driving the future of supply chains.

Speaker B: Everyone's talking about AI, but if you're running a warehouse, you're probably asking a much simpler question. How can it actually help me get orders out of the door faster? In this episode of Logistics Business Conversations, we strip away the buzzwords and marketing hype surrounding artificial intelligence and, uh, look at what it can really do inside a modern warehouse. My guest today is Scott Kramer from Infios, and together we're going to explore why the biggest opportunity isn't just about replacing people. It's helping them make better decisions when things don't always go according to plan. My conversation with Scott is about how AI is moving from a technology buzzword to a genuine operational tool. Yes, of course, there's some jargon, but it'll be wonderfully explained for those of you who, like me, struggle to keep up with the latest industry innovations. So if you are ever confused by terms like generative AI, agentic AI, or execution AI, by the end of this episode, I assure you you'll understand exactly what they mean and more importantly, why they matter. Uh, hi, Scott, and thanks for joining us today on Logistics Business Conversations. Now, I'm an experienced journalist who's, uh, interested in AI, but I kind of get mildly suspicious of all the hype around the technology. So, um, you've spent, according to your bio, more than three decades in supply chain technology, and you've seen it all, and now AI has arrived, and it's the new shiny big thing. So, uh, where does today's AI wave, uh, compared to some of the previous technology revolutions you would have, uh, worked through in your career?

Speaker A: Yeah, clearly, AI is in the top of that hype cycle right now. Right. Everybody thinks AI can solve everything. I do believe, though, it is going to be a transformational technology. You know, I don't want to say it's equivalent to the Internet or the equivalent of Microsoft Office, but I do believe that it's going to have a very transformational impact on businesses and specifically how warehouse operations are able to move and execute going forward.

Speaker B: As I walk around any, uh, logistics trade show today or attend a conference, every second stand seems to be using AI, claiming to be. So, as a customer, it's getting difficult to separate genuine innovation from clever marketing. So your team at Infios talks a lot about execution AI. What exactly does that mean?

Speaker A: So when we look at AI we're looking at it from the lens of how am I providing additional business value? You are correct. Everybody in every sentence or every third sentence of whether it be their website, their conferences and events, it's always about AI. But when we really start to break down AI, we see AI occurring kind of at four different levels. The first level that we see is what we call predicted AI. How do I try to interpret the data that I have based off of historical trends and say, this is what's going to happen in the future? Classically, you could think of this as a forecasting algorithm, but when we think of it, uh, in the area of execution, it's, I have a certain pick rate, I have a certain packaging rate. Is that truck going to get out on time today? Are all my orders going to get onto that truck? So first, the first lens that we look at is predictive. The next lens that we start to look at is generative, which is basically saying, I have some unknown data. I'm going to have the AI structure it and it's going to do something with that data and it's going to create something and give it to me more in a human term. So if I think about a slotting solution, historically with a slotting solution, I may have to look at a lot of data tables and trying to understand how did it re slot or how did it work with the warehouse, as opposed to coming back in a generative AI model and saying, I'm going to give you a natural language what my inputs were, what decisions I made, and then here's the resulting outcome of that generative process. The third area that we look at from an AI perspective is what we would classify as agentic AI. So this is what all the market's talking about right now. I'm m going to have agents running across the universe and solving world hunger. But we really see it as at an execution level. How do I take a lot of the manual discovery, a lot of the manual work that we do today, and really start to say, how can I turn a trusted AI agent, which we would call graduated autonomy, to allow them to start to do something? So as an example, if I have an agent that is looking at my inbound trucks and I see an ETA changed, I could have that agent maybe talk to another agent to say, okay, I need to reschedule my dock door because that truck's going to be showing up, as an example, four hours late. As a result, I may also want to redo my labor and my labor plan, and I may have to Even check and say, is there a hot item on that particular truck that now I need to cross dock? Right. So we start to say, how can agents start to help the environment in the execution world? And then the fourth area that we look at from an AI perspective is what we would really call conversational. So this is allowing an agent to go off and do a lot of the deep research for us that historically a human would have had to do. Right? Uh, if I look at a waving issue, I might have to go through and say, why didn't this thing wave? Do I have replenishment issues? Do I have lock control issues? Right. Do I have expiration dates? Just why didn't something wave? Where now in the future of AI, I can simply ask my agent and have a conversation with the AI and say, why didn't it, uh, wave? It will go off and do that 10, 15, 20 minutes of research for me in a matter of seconds and come back and give me an answer. Right? So we're really trying to say, what are the tangible business use cases that we can make really the management and the team leads of the warehouse much more efficient and give them access to data in a very, uh, rapid format.

Speaker B: So I think you've already helped me greatly with my comprehension of these AI subdivisions, shall we call them? So I guess as a warehouse operator, uh, they don't really need to know which one matters. They don't possibly really need to know which one they're buying. I guess they don't really need to understand the differences. But, um, therefore they should just maybe simply focus on the outcomes. Have I, have I read that right?

Speaker A: Yeah, yeah, I think you read that perfectly, Peter. Right. We're really focused on what's tangible business benefit they're get, they're getting rather than the technology. Right? I've been in the supply chain space for 30 years, and too often as supply chain, uh, companies, we focus too much on the technology, right? Is it one of those four types of AI? Is it an optimization? Is it a heuristic? Is it a microservice? At the end of the day, people just care about business outcomes, right? I'm very simple. At the end of the day, they want efficient operations and on time in full. Right? We've been chasing the same two metrics for the last 35 years, and AI is just the next iteration to help us to improve those metrics.

Speaker B: Right? Let's talk about a real warehouse. You know, a real warehouse operation. Let's say it's peak season. Well, labor's stretched. We Know it is because it is. Even when it isn't peak season, uh, there's a truck running late or um, a key customer suddenly, you know, change priorities as of course they're entitled to do. So how can AI be deployed by yourselves as a supplier to the industry, as a technology leader, to help operational teams, uh, deal with the reality of everyday life? Everyday execution in a warehouse.

Speaker A: Yeah, I think that's an important one as well, Peter. Right. Because we see needing AI to operate at the point of execution. Right. So we believe it needs to be tightly embedded into the WMS system. Yes, a customer could use Claude. Yes, they could use some of these other general purpose AI tools that are out there to maybe help them with some of their workflows and other things. But we believe that it has to be at the point of execution because the context of the data matters. I need to be able to understand the association, let's say as an example, between a license plate and order, the order lines and SKUs associated with that, the replenishments associated with that. So that's inherently what we can provide as a solution vendor that you wouldn't get from a general purpose. And allowing us to do that again gives then the operator, the team leads the data at the systems that they need to control the change. Right. So we see changing the world into more of a system of action rather than just purely classical exception management. Right. It used to be I, uh, would do all these real time data integrations which resulted in 10,000 exceptions because an event happened in the, in the ecosystem, like you were saying earlier, Peter. Right. Somebody drops a large order, that truck shows up late, five people call in sick on this shift, what happened? I as a user had to go and try to figure all that out. I start making a bunch of phone calls. I started doing a bunch of research into my, uh, user interfaces and spreadsheets and trying to interpret what I need to do where now we can assign a lot of that work over to these agents to help us understand the data and allow us to become decision makers rather than researchers.

Speaker B: Right. So AI is making the recommendations to enable the customer, uh, to make those decisions. Is there an element also of the AI executing its recommendations without, shall we say, consultation or without intervention from the customer?

Speaker A: I believe that will come in time. Right. And this is what we've been calling graduated autonomy. So at the beginning, you may not trust the agent, you always want the human in the middle. But then if the agent or the AI comes back and says, I'm 90% confident, then we allowed to start making those decisions, or if they're low impact decisions, allowed to start making those decisions. Maybe after a year of training, a year of learning, a year of it gathering additional context and additional information, we ultimately will allow it to automate for the more complex tasks. Right. If we think about simple tasks right now, I get an email, it might have an order change on it. I can allow it to scrub the email, understand that dates are changing or quantity is changing, and then send me a note or a notification through the user interface to say, hey, Peter recommended this order change, or Peter's requesting this order change. I've scrubbed the data, it's feasible. But I, as a human, I'm still making that interaction, say, yes, I can go ahead and make that change for Peter.

Speaker B: So, Scott, I often talk about the warehouse as being quite a controlled environment when I'm talking specifically about how AI harnesses data in the supply chain. The warehouse seems to be, in a way, the low hanging fruit. Because out there in the yard and beyond, there's a lot of variables. But that's maybe oversimplifying. I would say that warehouses can be messy environments, can't they? As you say, you mentioned, uh, full of exceptions, unexpected events, but also, you know, I'll throw in there, you know, the general clutter. You've probably got returns throwing in there as well, and all the vagaries of human behavior, which you mentioned as well. So warehouse execution, or AI in the warehouse is probably quite a difficult environment compared to say, you know, financial forecasting or customer service. So it's a complex task you're, uh, embarking on here.

Speaker A: Yeah, we see it as a very, very complex task. Right. And we are looking through the broader, what we call infeos, intelligence, supply chain execution. Understanding the interrelationship not just of the warehouse, but how do we connect out to our tms and how do we connect out to our oms? Because when we think about ultimately we need to have all three of those in a coordinated effort to get ultimately from order receipt to customer delivery. Right. So how do we make that entire process more efficient? Clearly, with the warehouse being a key component in that. Um, and you're right, there's a lot of things that can go wrong in the warehouse from day to day, from hour to hour. Uh, that AI allows us to help make, um, more efficient.

Speaker B: And when you look at the uh, priorities of your customers, you know, you get right out there areas where you'd think AI may not have, ah, a chance to improve, but of course it can I'm talking environmental, you know, sustainability. I'm talking about safety in the workplace. Could you quickly touch on areas where AI could help, uh, businesses in those departments?

Speaker A: Yeah, so at Infia's we're not directly doing anything with let's say safety as an example, but there's a number of vendors out there that are doing things like machine vision to make sure that people are lifting things in a correct manner, that they are doing the team lift when something is, is too heavy, that things are being stacked correctly. So there's a lot of areas outside of, on just say software that really get into the kind of the machine vision capabilities and some of the other AI, uh, capabilities that are also helping the warehouse today.

Speaker B: But I would guess, uh, a highly or very optimized warehouse by definition would also be an efficient one. Uh, so therefore less waste, less energy use and also I would imagine a more controlled, uh, environment would be a safer one too. So I guess those are the byproducts, uh, uh, of getting uh, efficiency, uh, through the warehouse.

Speaker A: Mhm. Yeah, I would agree. And if I just take one example, I mentioned slotting earlier. Right. It's just one example of if we can better slot the warehouse, if we can make more efficient user paths through the warehouse, if we can optimize the forklift so it travels 20% less in the course of the day. Right. All of those are getting to sustainability in the overall efficiency of the warehouse.

Speaker B: I would say that um, one of the great needs your customers will probably come to you for is uh, we need trust in this, we need reliability. Can you, can you guarantee that or how do you sort of build in those sort of AI trust, uh, recommendations into your AI?

Speaker A: Yeah. So we actually see AI, I'll say growing over the course of time. Right. The first thing that, that we see it doing is providing kind of that visibility layer, right. So it can sense the environment and can understand the environment. Then through training that we will provide out of the box, but then customers can continue to entrain, will allow it to start to decide on actions, right. It will allow us to give us recommendations, right. So whether that's a single agent or agents working in combination with each other, they can start to then decide and give recommendations. The third phase that we see this, this graduated autonomy working towards is acting, right? So this gets into your autonomous, allow it to start to make some of these decisions on its own. Then the last one is learning, right. We do see through these AI technologies the ability for them to continuously learn and improve. And that's why we're saying over time they may start with just information gathering and information sharing, but then ultimately move into this autonomous, uh, capability. And it will vary greatly on the complexity of the problem?

Speaker B: I imagine so, yes. And I imagine the sort of reliance on the human factor, on human decision making will never get taken away. I would have thought you're treating it very much as a tool rather than uh, something to replace, uh, for example, that famous phrase that everybody says it's coming to take our jobs. Well of course it's not. It's coming to make people's jobs easier, better, faster, uh, more efficient, more profitable, uh, whatever you throw at it. So, uh, Scott, do you think we can ever reach a point in, uh, a warehouse where warehouse managers, operators, um, could trust AI more than their own instincts, or will there always be space there for the human factor?

Speaker A: I think you're always going to have the human factor. Right. I think there's some amount of human decision making that AI is going to assist us with. Right. So we see it more as an assistant. For the simple menial tasks. Yes, go ahead and automate those. But for the highly complex tasks, you still need that human intellectual property, that human intelligence. Right. A lot of people fear that maybe the agents are going to replace jobs. But if we go back to the banking industry, ATMs came out, they said we're going to get rid of all the tellers. What's happened in the banking industry? They actually increase the number of tellers because they see it as a customer success or customer support, uh, capability rather than just saying the ATMs are going to replace the units. We believe the same thing is going to happen here. The managers are going to be freed up to make long term, better strategic decisions about how to make the warehouse more efficient and how to better satisfy their customers.

Speaker B: Scott, there's a great stat on your website saying that 79% of warehouse operators say that the execution speed is uh, what drives their advantage ahead of, of planning, say, or inventory optimization. Uh, I guess that's one of the great, great things that AI can do for them.

Speaker A: Yeah, I think when we talk about speed, Right. We've been on this journey again probably the last 30 years where it's all about real time data integration. And if I can only get the data faster, I'll be able to do things faster and I'll have higher, uh, customer service levels. But what we really find is it's the underpinnings are there, we have the data, but now I need to move from the data and have AI assist us so that I can actually get into an action. How do I take that data that I just got in real time, interpret that data, make a knowledgeable recommendation from it, and then act on that data. Right. So it's continuing to be how do we reduce that cycle time moving forward? And AI is just another thing that helps us to do that.

Speaker B: Now as a journalist, you know, I ask the questions and usually the question is what can this do? But uh, I'm going to flip that over now and say, what can't this do? So what are the limitations around AI here in the warehouse? What are the problems that customers often have that they uh, hope AI will solve, but uh, maybe just not yet, it just simply cannot yet fix?

Speaker A: Well, I see kind of two things that are happening. Uh, the first is they assume AI can fix everything. And the reality is you have to treat AI as a junior employee. They don't know your business, they don't know all of the workflows, they don't know how things work yet. Right. So you have to treat them as ah, a new hire or a new employee and allow them to learn over time. Right. They're not going to have all of the information from the very beginning. So this gets back to that human ultimately still needs to make a decision in the middle. The other thing that they don't do a great job of today is more so around us in the change management process. How do I start to go back to that trust and say, how do I start to trust the system? How do I get change management to work through the system? AI in and of itself is just a tool and it's how do we leverage that tool? I'd say the third thing that I see customers making a lot of mistakes on is they're treating it as a technology rather than trying to understand I have a business outcome that I want to achieve for that business outcome. Here's how I apply the technology, right? I think too often they're trying to say, I have this technology now, let me go find a problem. And maybe they're solving their own problems.

Speaker B: And I often hear around AI the uh, you know, some of those problems are centered around, you know, inputting bad data, bad warehouse data. I guess that the outcomes from that are just, you know, very smart but wrong. Uh, so I guess you got to with your customer when you're working with them, you've got to go right down to the nitty gritty, right down to the very beginning and see their data and make sure that's good to Start

Speaker A: off with, yeah, automating a bad process is still a bad process. Just execute it fast.

Speaker B: Yeah. Um, so one criticism of AI, uh, you could say, or one comment I've heard, is that it, you know, it's kind of all sort of lives in dashboards as I go sort of from conference to warehouse to exhibition. Um, anecdotally I'm now hearing, you know, alone, visibility isn't just enough anymore. So, um, is a future less about sort of knowing what's happening and more about how the systems actively respond to what's happening?

Speaker A: And again, I think this is why it's important that vendors like MVO start to say, how do we embed this directly into the workflows? Because you're right, it has started a lot with visibility. I can run a report, I can look at the data, I can understand what the system's doing, but I ultimately need to turn it into a collection of recommendations based off of some predictive capabilities that allow me to act on that data and understand what to do with it. Right. Whether that be autonomously or whether that be human in the middle. And I'm simply approving a recommendation or picking one of the three recommendations that may be provided back to me as the human.

Speaker B: I imagine, uh, there's some people listening to this who, uh, run warehouse operations. And, um, I understand why they may become somewhat overwhelmed by the amount of AI noise and terminology, uh, that's out there and the noise in the market. Where would you suggest they could start? What's the kind of. Is there such a a thing as a first use case that you'd recommend that they tackle with AI just to get them up and running?

Speaker A: Yeah, I don't think it's that cut and dry, Peter. I think it's really stepping back and saying, if they look at their operations, understand what AI is capable of doing, and then going back and saying, what is the one problem we want to try to solve? Is it, I want to solve a customer service representative type of an issue where I'm answering 10,000 emails and I want to maybe automate some of that capability? Is it as an example, that inbound ETA example that I gave earlier and saying, if I have this eta, best case scenario, my receiving manager gets the update from the TMS system or gets the update from the carrier. Best case, he maybe calls the warehouse supervisor and says, I now need to change labor off of this dock door and move them to, let's say, put away for a different dock.

Speaker B: Right.

Speaker A: A lot of those things start to break down in systems today because it requires humans to follow a standard operating procedure. When we think about what AI allows us to do, we can start to programmatically put that into the system with some deviations. Right. Classically, a lot of WMS systems have been very rule oriented, so I have to stay within a rigid and confined set of rules. When we think about AI, it allows us to be more interpretive of what we're trying to accomplish rather than putting in static rules. So I think it's really stepping back and saying, what's the use case I want to try to solve? And then saying, how can I apply AI into that specific use case?

Speaker B: I think you've kind of already answered what I was about to ask, uh, which is, is there sort of a common mistake that companies make when they introduce AI into their warehouse operations? Perhaps the answer is bad data, but are there any other sort of areas where people trip up early doors and need rectification early on?

Speaker A: Um, yeah, I think it is the bad data, but I think it's also combined with, ah, as we kind of mentioned earlier, they have a process and they just want to automate a process rather than maybe stepping back and saying, how should the process change? If I have a more rapid ability to understand the data that's available to me, right. Would I come out with the same decision? Would I come out with the same business process? And then starting with the end in mind, right. I want to have this outcome, therefore I need to back in and say, here's how I apply one of those four types of AI's capabilities, uh, I talked about earlier.

Speaker B: Now Scott, you've watched, uh, supply chain technology evolve over, uh, uh, what do we say, three decades or so. Um, so I'm afraid that makes you highly qualified to answer my crystal ball question. You'll be glad to know. Um, so when you and I are sitting down again in what, five years from now, um, maybe in the Bahamas with a cocktail in our hands, I don't ever know, um, what will have changed, would you say most dramatically inside warehouses in the year? Uh, let's do the maths. 2031,

Speaker A: what's next in logistics? We're asking the experts in our crystal ball segment, Industry leaders share their predictions for the future. I think the biggest change is going to come in at the management tier, right. I think your team leads and managers are going to find, not that they're going to have free time, right, Because I think that free time is going to get reoriented into other strategic decisions and other strategic activities. But I think the amount of manual effort that they did on research and data analysis is going to tend to go away or on repetitive tasks will start to go away, freeing them up to really start thinking more strategically about the business.

Speaker B: Well, I would say, finally, Scott, what would say is the biggest misconception that the logistics industry, uh, currently has about AI? And what are the questions your customers come to you repeatedly that suggests they don't quite understand what's happening around them?

Speaker A: I think the biggest one is kind of what you asked a little bit earlier, Peter. They're trying to understand how do they leverage this technology. They've been given an amazing tool and they're still trying to figure out how to leverage it and how to use it. Um, and I think that's where having these types of podcasts, having these types of conversations with customers, has really been insightful in helping them to understand where are the right outcomes, where's the right way to apply the technology.

Speaker B: It feels like the industry's undergoing quite a step change in technological advancement. And, uh, I also sense you're very, very much at the forefront of that. It's a very exciting time to be in logistics and supply chain technology right now, isn't it?

Speaker A: Uh, oh, it's a great time. I am at heart, ah, a supply chain person, but I'm also a technology person. I've always kind of stood between the intersection of those two worlds. And I think, as I said earlier, AI is going to have a pretty significant impact, uh, on how executions work moving forward.

Speaker B: Scott, it's been great having a conversation with you today. Thanks a lot for joining us on logistics business conversations and hope to meet you in real life at some point.

Speaker A: Yeah, I'm looking forward to my cocktail in the Bahamas.

Speaker B: Peter, I really enjoyed that conversation with Scott. He's, uh, helped me clear a few things up in my mind that I was just getting a little bit tied up about, and I hope the same for you as well. The takeaway there, I think, is AI is only as good as the information that's fed into it. I loved in particular the little bit when he's talking about treating AI as a new employee and feeding and training it. You know, it's really exciting what the future holds, and companies like Infios are at the forefront of that and they're going to be, uh, seeing great changes coming in the, uh, warehouse of the future. I'm really excited for that. So all it leaves is for me to say thank you very much indeed for listening to this episode of Logistics Business Conversations. I urge you to like and subscribe because that really helps the visibility. And also visit our website where you get a whole back catalog of these logistics business conversations going back for many years, shall we say. So thanks again and look forward to talking to you at, uh, the next one.