Turning Warehouse Blind Spots Into Real Time Intelligence With Dexory

Speaker A: Your agentic AI might not be secure even with real time data and proper guardrails, but denodo makes sure your business has every avenue covered. By placing all your data platforms under one AI data layer, your business can reach semantic consistency safely and securely. So get your agents on the same page by visiting denodo.com and you can learn more about how to start trusting your agents to make business decisions. What if your warehouse management system says a pallet is present, but the physical warehouse tells a completely different story? Well, a company called Dexary has built autonomous robots that are capable of scanning inventory, uh, at heights of up to 18 meters. And they're processing between 10,000 and 12,000 palette locations every hour. And I struggle to remember where I left my keys. So 1 billion scanned locations feels like something incredibly cool here. So today I'm joined by the co founder of Dexary, who leads the company's commercial strategy and product roadmap to. And together we're going to discuss how physical AI is creating a continuously updated view of warehouse operations and why collecting data without prioritizing action can quickly overwhelm teams. But I want to explore the world of robotics, how they can reveal lost stock, wasted movement and unused capacity. So if your supply chain systems cannot see the physical reality, how reliable are the decisions built upon their data? This is just one of many questions we'll explore today. But enough from me. Let me introduce you to my guest right now. So, thank you for joining me on the podcast today. Can you tell everyone listening, uh, a little about who you are and what you do?

Speaker B: Sure thing. Thanks for having me. Niels Ginga. I'm one of the three co founders at Dexary. Uh, and, uh, I run everything that has to do with commercial and product roadmap and strategy for the company. Um, been doing that for quite a few years now. Um, and, uh, yeah, happy to tell you more about what Dexary is. Um, yeah, later on.

Speaker A: Yeah, I'm looking forward to it as well, because one of the reasons I was excited to get you on the podcast is you set off my tech Spidey sensors when I learned that Dexary has built an autonomous robot that can scan inventory at heights of up to 18 meters, which is incredibly cool. But, uh, before we even get into the tech side of things, what warehouse problem led you to that design and what can it detect that maybe conventional stock checks often miss now?

Speaker B: Yeah, so I, I will take a, well, a step back into probably more than a decade now because we've been working together as a founding team for uh, over 10 years. Pretty much started with this idea of using autonomous robots to collect data from the physical world, to kind of digitize the different physical spaces. And initially, uh, we created something for the home. So you kind of keep an eye on the house while you're away and, like, uh, capturing what's, what's happening there. We then moved into retail stores, which was going really well, looking at, kind of taking people around to different product areas, uh, but also, as we were doing that, keeping an eye on the shelves in the stores to make sure stock is there. Uh, and obviously if something kind of is low, that it can be replenished so people can buy as quickly as possible. Um, and that was going quite well, um, before the pandemic. But obviously 2020 came, um, and it was a bit of a rocky kind of start to the year with all the lockdowns and of course, our market disappearing overnight because all the stores closed, so no one obviously was going to physically buy anything. Um, so it was during the pandemic that we kind of first got a glimpse of. There are similar problems as you have in a retail store, but inside warehouses. Whereas people kind of know what comes in and what goes out of a warehouse, but what happens with the goods and the people and the machinery inside is a little bit of a black hole. Like, you don't have GPS trackers to understand where stock goes. Um, there's obviously, um, a lot of chaos, especially during the pandemic, because you could only get certain number of people in. You had to get a huge amount of orders out because everyone was ordering online. So it's a very, very kind of chaotic space that needed a lot of digitization. Uh, so we took a step back and thought, like, okay, can we do what we've done before in the stores inside the warehouse? Uh, and that's kind of how we stumbled upon this whole kind of area of stock and visibility inside the warehouse. So coming back to, I mean, more recent days, uh, in 2023, is like when we launched the solution for warehousing and never looked back. But the actual kind of question you had is like, yeah, what's the problem? And the problem that we address is you kind of need eyes across everything inside the warehouse, because most of the decisions and most of kind of the systems record what they think should be there, uh, where actually the reality on the floor is very, very different. Um, so we kind of provide that visibility inside the site from top to bottom, hence the whole 18 meters point, because you do have goods that are sitting at 1820 meters up in the air, uh, which with human eyes you can of course, even see to kind of go all the way up there or bring the goods down and have a look at them. Um, but also, um, we are pretty much digitizing these spaces at an insane speed. So it's not just the fact that we capture information that humans can't because of height or because. I mean, yeah, you can go and scan a barcode, but you don't. Well, yet you can't take a picture of it with your eyes. You don't understand volumes and shapes and sizes just like by blinking. Uh, so we capture a lot of data you can't normally do with people. But secondly, we do it at crazy speed. So that's what I'm saying. Like, we do about 10 to 12,000 palette locations an hour. Uh, which means, like, in a few hours you kind of get a full view of exactly what you have there. We continuously kind of update that, uh, over and over as the day kind of goes. So, um, yeah, the problem is the visibility, uh, that was lacking. And then once you have that visibility, we've evolved so much because we can do so much with that data. Uh, once you have it to start

Speaker A: with, it's incredibly cool what you're doing here. And just to help listeners understand the scale of what we're talking about, your robots have already scanned over a billion warehouse locations. So I've got to ask, when you're scanning that volume of data, what has it taught you about the gap between what companies believe that is in their warehouse and what is actually inside there? I suspect you've got a few stories.

Speaker B: Oh, loads of stories. I mean, from the very simple things that, as we normally go in, it takes us up to like five to seven, seven days to get everything up and running. So once we kind of get a customer on board, it's very, very quickly to actually start getting them data. Um, and, uh, I have so many examples. Whereas, like, with those first scans and the first time to digitize the space, we just find so much stock that they thought that they had lost or that they written off and they had to pay for. Yeah, that's kind of the most basic thing. Of course, you're going to find things that you didn't even know existed. And I think the biggest number of items we found is probably somewhere around, like, £1.5 million of stock, which obviously is quite a big am, uh, depending on the site. Um, so that's kind of like number one, but number, um, two is like, as you continuously get updates on this stock, because, like I said, it takes us a few hours to do it. And then the robot charges a bit, goes out again and captures the data again. It's more like the trends of what happens in between the scans as well, which I think is always very interesting. Um, so you can kind of see how many times something has been touched or moved around. And then our users are like, wait a second, why do we keep moving pallets around for nothing? Like, you should only touch something once as it comes into the warehouse in seconds, as you kind of get it out of the warehouse. But there's a lot of kind of random movements that are happening inside because they're maybe not respecting the, I don't know, basic compliance rules, like the weight on the racking or something like that. So then they have to kind of correct it. So there's a lot of kind of wasted resources to do things that don't necessarily have an immediate business impact. And that's something you can track once you have that data continuously. Uh, but there's also elements around how you utilize space. Like, we have a lot of customers that are saying, well, our site, our warehouses are telling us they're like full capacity. And then when you actually kind of start scanning, it's like, well, they're not really at full capacity. And actually if you shuffle these things around, you kind of create another 10% of capacity. So there's a lot of elements on the data that kind of, again, you kind of have to have that ground truth. And then depending on the sector and the customer, there's so much you uncover by crunching it in different ways and comparing it with different systems, and, um,

Speaker A: real time visibility will sound incredibly attractive and for all the right reasons. But if one thing we know about data in any industry, and that is collecting data only can create value when somebody goes out there and acts on that data. So how are you seeing warehouse teams using that intelligence during an ordinary working day to maybe reduce errors, delays, or wasted capacity and get that return on investment? What are you seeing here?

Speaker B: Yeah, so, um, when we started three years ago, I think the number one thing we're trying to do is just get the data to the people. Uh, and then what we realized quite quickly is like, well, you're inundating them with just a lot of information. And to your point, it's not just that you have to kind of find time to look at the information, but if that's overwhelming because you're seeing, I don't know, like 50 things on red, there's like an immediate reaction. So I'm just going to close my laptop and go awake, but don't want to deal with this. Right? There's too many problems to deal with. Um, and I kid you not, the number of sites that we went live with and the first kind of scans highlighted so many problems that they're like panicked and like, oh, we don't have time to deal with all these problems. Well, if you don't deal with the problems, they're only going to have a knock on effect on everything else you're doing. So clearly you probably do want to address them. But um, what I'm trying to get at is like we learn very quickly from our users that we have to help them um, sort and synthesize and also kind of like prioritize what they should be working on. So going back to the points, like it's not just that we're giving them the data, um, but we are turning that data into proper action, uh, so that we can then um, uh, help them understand what they should be doing about it and how they should be um, utilizing it immediately. So within our DEGS review platform, which is a digital twin, the number one thing that you land on is like today these are like 10 things that you should be addressing rather than here's a lot of information.

Speaker A: And um, of course physical AI combines machines operating in very real environments with software interpreting everything that they encounter. But what is it that makes uh, a warehouse harder for AI to understand than maybe a, a controlled lab or a digital system or something in the corporate space? What kind of challenges are there?

Speaker B: I think people forget that a warehouse is like a living, breathing environment. I would say there's just so much happening, uh, at different times. I mean, I think there's this perception that you kind of put something on a rack and that's it and no one kind of looks at it anymore. But that's obviously not the case. You have people that are coming in to pick orders, that are moving things around, that are cleaning. There's just so much happening every single hour that without having that continuous visibility in it, you will start making some assumptions based on synthetic data or trends that just actually don't translate into the real world. Um, and there's also things like, um, I don't know, for instance, you kind of understand routes and ways of kind of say navigating a warehouse. You kind of have an initial CAD file of what that looks like when the racks are put together and the walls were built. Uh, but it's kind of Surprising to see how many things change literally, like on a daily basis, like something else gets put in an area, or like an area gets destroyed or flooded, or the environment itself changes so much, um, continuously that again, without having that ground truth, there's going to be a lot of failures when it comes to physical AI, because you're just going to assume, yeah, I can kind of navigate through that area, but actually the people on the ground kind of know that you can't. But there's no way to kind of get that out of their minds and into a digital system. So there's a lot of, kind of rules on the ground that are like, unspoken, unwritten, that you can only kind of capture by continuously kind of scanning these places. So I think that's kind of the number one failure that I see could come from, um, the warehouse space is like, people assume that they're very consistent where actually the reality on the ground is very different. They're very, very dynamic.

Speaker A: Um, businesses can obviously automate a poor process and produce mistakes so much faster, even at machine speed. So for anyone listening before they begin investing in robotics, what should a warehouse operator be examining across their, their workflows, their data quality, layout and systems integration, and workforce readiness? What do they need in place before they go on that journey with you?

Speaker B: Yeah, I think it's a, it's a great question and I will be honest to say, like, I don't think they have to wait and try and get everything right, uh, before they start, because that then kind of takes them into, um, another area which is like a lot of wasted time just doing things the way they were done before and obviously still kind of like not getting that level of efficiency. But I think that data accuracy point obviously is the first step. Um, and it's a bit different if you're thinking about digital AI or physical AI. This is what I'm saying, because from a digital AI perspective, um, data exists in many of these cases. Like you kind of have some sort of a system or like an ERP or a transport management or something where you have that data, but the physical data, which you can't really capture at the moment, that you have to think, okay, how can I start capturing some of it? Whether it's by asking people to report on things or adding some cameras or having a solution like ours to even get, uh, them going. Um, so I think it's, yeah, trying to understand, like, what data do I have and where and what kind of tools can rely on it already. Um, but the other thing that I always tell people is, um, it's not necessarily about readiness per se, but it's more like trying to understand what problem we're solving so that you don't just bring technology in for the tech's sake, you bring it to solve a particular problem. So then if you start with a problem in mind, whatever tool you might bring to solve it can differ so much. So a simple example is like, okay, I want to increase how many, um, orders I get out the door for that. You can look at a variety of different solutions to help with the picking. It's either autonomous trolleys or autonomous forklifts or little robots and they that kind of help, uh, fulfill the order. Um, so narrowing down that problem to, well, what exactly am I trying to achieve? Like, I need to get small items packaged really quickly. They're all in this particular area of the warehouse and I still have humans that can work with the robots. That's like one solution that you're looking at. If you're thinking about like, I want to just completely automate everything, that's a different type of solution. So I feel like narrowing down that problem to begin with is probably the most important thing they need to do. And then the rest is going to follow.

Speaker A: And there'll always be nervousness when we talk about adding AI and robotics into an industry like this. So for any skeptics listening or people concerned about their job, et cetera, what happens to those traditional warehouse roles when robots take over the repetitive inventory scanning and what human capabilities become even more valuable? How should employees maybe be preparing existing workers for a new level of responsibilities?

Speaker B: Yeah, I think the, the number one point is, and something that we're hearing across the board from all over the world, I should have said, like, we, we have customers from like, Australia and to the Middle east and to like, Europe, the U.S. canada, Mexico. So like pretty much kind of a global, um, uh, coverage at the moment, uh, and everyone is telling us the same thing is like, well, there's not really a big risk here because we can't really find people that want to work in warehouses anymore. So I think the number one problem is actually finding the workforce to begin with because I think, yeah, it's not the most kind of glamorous environment. There's also a lot of competition for the workers. And you have the likes of Amazon obviously kind of coming in and taking a lot of that workforce by just paying a little bit more. So there's a massive kind of competition for heads to begin with. Um, but I think for those that fear that the role will be automated. We have so many examples where that kind of whole team shifted in responsibility. So rather than having to do very basic, they call it walkabouts, the warehouse, to try and kind of capture, um, what's wrong and just literally kind of with a handheld scanner or like pen and paper, just note things down. Comb boxes. Obviously that's a very kind of boring and tedious task. So once you kind of take that away and you kind of provide them with insights and data, then that kind of changes them to be more like in a data analyst space and actually figure out, okay, what can we do about it? Like, what decisions can we make? So we have, uh, loads of examples from sites where that's happened. And, uh, the teams on the ground have been so, so excited to just be able to forget kind of the worst parts of their job and actually, um, focus on the stuff that gets them excited, which is like, okay, how can we make things more efficient? Um, so I think, yeah, that mindset of growth and trying to figure out, um, okay, once we have the automation in, um, how much time does that kind of, um, free up for me so that I can do something else? Um, is something we're seeing across the board.

Speaker A: And the tech industry famously has a diversity problem. There's been a lot of ground made, but there's still so much more that needs to be done there. And I think, especially at a time where diversity of thought is so much more important now, especially in solving increasingly complex problems and the world of robotics, if we just take this area for a moment, it still struggles to attract enough women into the industry. So, uh, someone right in the heart of this space. Have you ever encountered any barriers that have impacted you personally? And what do you think, or what practical changes do you think schools, employers, investors and industry leaders right across that entire chain could actually make to help improve, widen participation?

Speaker B: Yeah, I think it's changed so much since I started, like, over 10 years ago. First things that I was doing at robotics, um, and I remember kind of going to pitch competitions and everything else as we're starting the company and pretty much being the only woman in the room or maybe like one other. Right. So, um, um, and also kind of being a little bit frowned upon by everyone else in there, being like, oh, look at them. Right. So it's just like a bit of a different type of environment 10 years ago, whereas now, if you look at it, there's just so many examples of, like, incredible women, obviously building some fantastic companies in the physical AI space. Um, so I think the number one thing that everyone can do, not just in academia and so on, is kind of keep bringing these examples front and center and talking about them because that then inspires kind of the new generation and helps them understand. Like, yeah, I can definitely do that. It's not just a field that I wouldn't normally go into because I don't have role models to understand that it is a possibility. So the one thing I keep saying is as people give examples of success or career path and everything else is kind of using um, um, these role models that already exist, um, so that you can kind of get people to understand like there you go. Like that's a normal thing to do is kind of have a woman, uh, obviously like either like leading a robotics company or building physical robots. Like it's actually a space. And I think with AI, actually I was talking about this with somebody else last week, um, uh, in the context of like what can you do with AI tools? And I think if anything it's leveling up things so much because now with the accessibility of the tools, you can code from scratch and then kind of build things that before you kind of needed a four or five year computer science degree to do. Right. So it's kind of helping to level up the game. Even if you did not study that beforehand, you can now kind of go into tech areas that um, would have required a lot more effort before. So you kind of have to reinvest in the training and in yourself, whereas now you can kind of jump straight, straight into the building. So those people that have incredible ideas, regardless of gender or background, uh, can actually get into building them and bringing them to market much, much easier, which is obviously like we're seeing is leveling the field a lot for, for the different groups.

Speaker A: Fantastic. Thank you. And I'd love to give people listening further valuable takeaways here. If we have a supply chain leader that's maybe assessing physical AI today, that's why they're listening to this episode. Or what operational problem do you think makes the strongest starting point? That uh, low hanging fruit. And what results should they measure? And the old saying in it is you can only improve what you measure. And also what warning signs would suggest a robotics project is unlikely to deliver anything, uh, you can share here from your conversations you've been having?

Speaker B: Yeah, there's a lot of different bits in your question there, so I think I'll try and kind of break it down a little bit. And um, I think going back to the point earlier in the sense like what should you be thinking about is going back to, especially in the warehouse itself. It's actually going back into the warehouse and physically observing things that are happening, um, around spending a day to just walk in the warehouse, seeing what the teams are doing and uh, what opportunities there are to drive more efficiency. Um, and secondly would be asking the teams themselves as well, where do we see the, we have bottlenecks, um, like how can we then improve some of those throughput or whatever, um, um, kind of KPI they have. So yeah, just going back to the drawing board and just observing will then highlight probably a few areas that um, would benefit from any kind of tech, not just kind of automation. Um, and then once that's determined, going back to my point earlier, really figuring out what is the actual problem that we're trying to solve to then determine what type of solution might be the best one. Uh, and thirdly, as I've discussed this quite a lot is getting that ground truth and that data, uh, to be as accurate as possible regardless of the project, it's going to make sure that that project that comes in will deliver the best outcome. So we talked a little bit about picking robots. Um, if you send a picking robot to a location and there's no stock there or the stock is not in the right quantity or it's damaged, that's going to be like a missed opportunity for that robot. So then it's going to have to go and wait for another task to be assigned or it goes to what we call the doctor, going to have to wait for a human to look at that, um, order in itself. So the automation is not as efficient as you thought it would be, but not because of the automation's fault. It's because of some of the ground truth being wrong and you're asking it to do something impossible. So again, this is why it's very important to, once you nail the problem down, to understand, okay, how much do we know and where do we still have gaps? So make sure that the project's going to be as um, uh, effective as possible. And I think in terms of what I've seen where stuff doesn't land is if it doesn't have enough kind of support internally. And by that I don't mean like necessarily kind of buying from everyone working on a site or anything else, but at least having somebody that owns the uh, project success as their kind of responsibility at the end of the day is having a sponsor internally or a champion that is really kind of working with the different teams to help them understand going back to the Basics like, okay, why are we doing this in the first place? How is this, this going to help you? Why is it important for you to embrace it? Um, and then kind of being there to champion for the solution. If you don't have that champion, we just drop something in and you hope people are just going to run with it. Usually that fails very, very quickly.

Speaker A: Well, you've been on a phenomenal journey and I think first put you on my radar. There's an article in the Times last year and Dexteri raised, I think it was something like 165 million to fund expansion across Britain and the US. So fast forward to present day. Where are you now? What excites you about the future and everything that you're working on? And what can people listening expect from you guys in the months ahead?

Speaker B: Yeah, ah, there's always a lot going on. Um, and I think in the context of, again, physical AI, it's now like making robotics be a very hot topic across, uh, of customers as well, but even kind of investors in the tech world. Uh, because when we started about 10 years ago and we're telling people that we're making robots, everyone thought that we are a bit crazy because it's hard to make hardware. Right? Whereas now people realize, okay, no matter how much you do in the digital world, something has to happen in the physical world that can actually kind of influence, uh, the reality in society. Right. So it's great to see that the world's kind of caught up with it and that's now a very hot topic. Hence obviously why it helped us raise the money and people kind of getting behind us. Um, but, um, I think the number one element is, as I was saying earlier, we started with collecting the data, but now we're finding out more and more things we can do with it to really bring value to our users. Um, and this is kind of what we're very, very actively working on, pretty much kind of nonstop. It's like, what else can we extract from this data? Uh, where can we bring value to different teams across different sites that they have, um, and really kind of embrace, well, digital and agentic AI as well. On top of that, physical data that we collect also kind of integrations with a lot of different, um, let's say actors in the field. So, uh, obviously sometimes the action that we suggest is taken by humans, sometimes by the robots. So there's a very interesting kind of element there on how you can orchestrate those resources, um, inside the warehouse. So, yeah, it's extracting more and more from the data and that orchestration piece. Um, and I already kind of mentioned we keep expanding globally. We're growing a lot in different, uh, markets even without having teams on the ground. It's quite an organic growth with, with our customers because we now work with some of the biggest, um, logistics companies in the world. And we're lucky enough that, yeah, they do have a global footprint and they do have sites all over and they want us in it. Uh, so that's going to continue to grow as well in terms of our footprint.

Speaker A: Well, thank you so much for sitting down and sharing your story with me and everybody listening today. And again, anyone in this industry wants to find out more information about physical AI, robotics, what you're doing in this space, where would you like me to point everyone listening?

Speaker B: Yeah, I would say, like, follow us on LinkedIn. We share a lot of stuff on LinkedIn and of course like our website. But, uh, I think, yeah, we have a fantastic community on, um, um, LinkedIn. And, um, I think, yeah, we're trying to always bring a lot of things that we just discussed, like what are we seeing in the ground, like what's best practice, like what's our view of the future? Um, and also, like, we talk a lot about some of the other solutions in the field as well.

Speaker A: Fantastic. Well, I will add links to the Dex Re website, LinkedIn, your social channels, et cetera. I know people to follow on there. You'll get. That's probably the quickest way of finding out more about those new announcements as they drop and also get in touch with them as well. There's so much going on here and some big gains to be made, some big opportunities. But thank you for sharing your journey today. Really appreciate your time.

Speaker B: Um, thank you for having me.

Speaker A: I think my guest story shows why warehouse automation must begin with a clearly defined operational problem. Buying an impressive robot and searching for something useful to do with it, of course, would be an expensive way to decorate a warehouse. But the value here that we're talking about comes from connecting physical AI with accurate data, measurable outcomes, and, um, people who understand their work. And then Dexry's robots can scan thousands of locations every hour. But that, uh, information must be prioritized before, before it can become useful. And giving a team 50 red alerts might create panic. So giving them 10 actions that matter every day could improve how your site operates. But I also loved her, uh, workforce message there, especially around automation, how it can remove repetitive stock counts but allow employees to spend more time interpreting information, solving problems. Overall warehouse performance. But over to you. Which repetitive, uh, warehouse task could you automate first? And if you could, where would you ask your employees to do with that time that has been returned to them? Techtalksnetwork.com that's where you can find me. I'll be back again real soon with another guest, but keep your stories and, um, experiences coming over. Speak to you soon. Bye for now.