- Welcome to another episode of "Supply Chain Frontiers", the MIT CTL podcast where we explore the trends, technologies, and innovations shaping the future of supply chain management. I'm your host, Mackenzie Berry. Today we're diving into CTL's latest research on warehousing and e-commerce through the state of AI and warehousing report, a collaboration between Mecalux and the MIT Intelligent Logistics Systems, ILS, lab and the 2026 State of Omnichannel Supply Chain Report. Joining us are Dr. Matthias Winkenbach, director of ILS. Alejandro Gonzalez, software business unit director at Mecalux. And Dr. Eva Ponce, director of the MIT Omnichannel Supply Chain Lab. So glad to have you all on the podcast today. Can you all introduce yourselves? - Thank you for having me. I'm Eva Ponce. I'm the founder and director of the MIT Omnichannel Supply Chain Lab. In this lab, we are working on developing and creating new insights on how e-commerce and omnichannel fulfillment are transforming supply chains. We work with innovative companies and we also apply a cutting edge research techniques methodologies. We work on core areas. Our core areas are omnichannel fulfillment, identify new trends in warehouse automation, omnichannel returns, and also the role of technology and AI in e-commerce and omnichannel. - I'm Matthias Winkenbach. I'm the director of research at the Center for Transportation and Logistics at MIT and I'm also the director of the Intelligent Logistics Systems Lab at the center. That lab focuses on a variety of topics. Traditionally, we've done a lot of last mile logistics research and lately, we've expanded into the warehousing space and as part of that expansion, we started collaborating with Mecalux, who's currently one of our major research partners. And so I'm particularly excited to have Alejandro join us on this podcast again today. - Hello everyone. First of all, I'm very excited to be here. Thank you for inviting me. Once again, I'm Alejandro Gonzalez, software business unit director at Mecalux and in my role, I'm responsible for leading the global business strategy, driving growth, and overseeing the sales performance of all our software solutions portfolio worldwide. - So glad to have you all. To get started, for Mecalux and ILS, we had you all on an episode when your research partnership had just begun. What can you share about what you all have accomplished together so far? - Last time Alejandro and I were on this podcast, we had literally just launched this partnership and since then, we've come a long way. We've moved from talking about the research that we wanted to on a conceptual level and we've come towards actually deployable AI solutions, AI tools. So that was the focus of this partnership from the get-go, not doing purely theoretical machine learning research, which is obviously required as a foundation of what we do, but practical AI solutions that solve real logistics problems. Just recently we completed the work on several of these first year projects with Mecalux. To highlight one of them, we just came back from Barcelona presenting to their management team a tool that we call GENESIS, which is basically a simulation-based AI-driven tool that helps any company manage and optimize inventory across multiple warehouses. What we're basically doing is we're using something called a genetic algorithm to evaluate a lot of different potential scenarios that could or could not pan out. And then across these scenarios, the tool actually optimizes inventory, stock levels, replenishment, timing, and the like. Why does that matter? A lot of companies, especially in a omnichannel context, face this problem. If you wanted to solve it really well, it would take a lot of time to do so computationally, a lot of resources will be consumed and we've basically cut that down to just a couple of minutes of running a scenario. So that really enables a company like Mecalux or some of their customers to do tactical planning, scenario planning, not just purely theoretical analysis. So that's really good to see that even within the first year, we came to something that is not just a nice little academic paper, hopefully, that also comes out of this partnership, but also something that could be turned into a solution that is gonna be deployed in the real world. - From Mecalux's perspective, we're really satisfied with what we've achieved during the first year of the collaboration. As Matthias was saying that the partnership has already delivered several applied research projects that are strengthening key parts of our software portfolio. For example, we are strengthening our distributed order management solution by using genetic algorithms, as Matthias just mentioned. And the idea behind this is is that we want to be able to optimize how inventory is positioned across the supply chain network. And in simple terms, what we wanna achieve with that is to help companies place the right stock in the right location, but more intelligently. And at the same time, we are making order fulfillment smarter through machine learning based orchestration so the system can dynamically decide the best way to execute and fulfill orders. And I wanted to highlight also on the automation side of things that we are improving also how AMR fleets operate in complex environments using advanced AI algorithms to better coordinate and optimize our own fleet management system. But beyond the technology itself, I think that what really matters to us is the practical impact because nowadays, we as Mecalux, we operate thousands of facilities worldwide, and so we see, you know, firsthand, what are the operational pressures that our customers are facing day to day. And this is why this kind of partnership in the end led us to one clear objective, know that we wanted this collaboration to help us address those real logistics challenges in a smarter and in a faster way. Therefore, what makes this partnership especially powerful, I think that it's the speed because instead of research staying in the academic world, we are moving quickly from ideas to pilots and from pilots to real customer deployments. - And can you share for those who don't know what a genetic algorithm is? - So that's hard to explain in just a few minutes, but I'll try. So genetic algorithm, the name comes from biology, as you might imagine. So the idea is basically you have a pool of potential solutions to a problem, which may or may not be good or bad. And then like in a gene pool that gets crossed over as a population propagates, those solutions get mixed. So you basically take a part of one solution and take the other part of the other solution and you recombine them to create, in a way, child solutions. And you do that over many, many generations to generate a very large pool of solutions, some of which may actually not even be feasible anymore. But the more you do this, the more likely you are to find better solutions to the problem. And if you do this systematically and at scale, you will eventually find near-optimal solutions to the problem. - So trying a bit of survival of the fittest in a sense? - Exactly. - And for Mecalux, why did you all decide to conduct the research with the ILS lab to make the State of AI in Warehousing Report? Why is that important now? - Well, over the past few years, AI, and especially generative AI, has dominated the headlines. That's not news to anyone listening today, but what we've really seen is that our customers' questions have really evolved because a few years ago, they were asking, "What is AI?" Today, they're asking, "Where should I invest? What's actually proven? What tangible results companies are seeing? And also what kind of return on investment can I realistically expect?" So that shifts is exactly why we felt that this was the right moment to conduct this kind of research because we wanted a fact-based, data-driven snapshot of AI maturity in warehousing and something that goes beyond marketing noise. And more specifically, what we wanted is to understand where AI is truly being deployed and what measurable impact companies are actually achieving. So for us, the report is about bringing clarity to decision makers and I personally think that it gives logistics leaders grounded view of where AI is generally creating a competitive advantage and also where it's still more a promise than a reality. - Absolutely. I think everyone's eager to look under the hood of AI, so to speak, and see how can it really be applied. For Eva, your lab conducts the State of Omnichannel Supply Chain Survey annually. Why do you conduct it annually and what did you do differently this year? - Yes, so the omnichannel survey captures how organizations across different industries are navigating omnichannel complexity. So it provides benchmarking tool for industry leaders to identify the challenges in omnichannel, what other companies are facing, and more importantly, which techniques, which tools they are incorporating or deploying in order to face these challenges. We conduct it annually because this help us to track evolution. So we are able to identify patterns and see how some of the tools are much more material now and some of the tools are coming now to the field. So it help us to identify operational pain points. What we did differently this year is to dive deeper into the role of AI. So technology and automation has been always core for our survey. A couple of years ago, we started asking questions about are you using AI? In which functions? But this year, we try to ask specific questions about what are the specific AI tools you are incorporating and in which functions are you incorporating these specific AI tools? Are you incorporating machine learning techniques? Are you incorporating gen AI, LLM models? And the functions, we have in deep dive a little bit our custom experience, demand forecasting, inventory management, warehouse management, and transportation management. - And can you share all the industries that omnichannel touches for those who may not know and what kind of respondents you're seeing in the report finding? - Yeah, that's a very good question. So in terms of industry, we have a very good representation for food and beverage, electronics, furniture, fashion is also very well represented, as well as beauty and pharma. And it makes sense because these are kind of the industries that more actively are implementing omnichannel strategies. We also have, as part of the survey, different actors as part of the supply chain. We have a huge representations of retailers and manufacturers, CPG companies, but we also have important representation of third party logistic providers who are the companies that are working with many retailers in providing the fulfillment services. So in that sense, it's very broad in terms of the industry as well as in terms of the different actors in the supply chain. - Let's dive into the report findings, particularly around AI and automation. So the State of AI in Warehousing Report finds that adoption of AI in warehousing is widespread, as we can imagine. Can you all share how widespread this adoption is and more specifically, how and to what extent warehouses are using AI? - Yeah, so that's one of the key findings of the report that in previous years, I think AI adoption was still somewhat experimental and now it's really become mainstream based on the responses that we got. We actually only had about 1% of all of our respondents who said they are not already using or at least piloting AI. And 82% said we've increased the use of AI in the last year. What's also interesting to see is that the depth of adoption has increased. So we are actually beyond a stage where people just do one pilot after the other. People are actually looking into scaled deployments of usable tools on an operational level. So for instance, it varies a little bit, but the majority of our respondents said that they are already using AI in somewhere between a quarter and three quarters of their warehouse operations. They also said that about 60% of our respondents are already responding kind of advanced or full automation maturity. So we're not just talking about very basic AI technology, but actually pretty advanced deployments already. If you look at what the most common use cases currently are, not surprisingly, most of them are still focused on improving relatively easy to measure metrics like cost per order or service level. So we're seeing a lot of applications in fields like inventory optimization, automated picking, optimization of routes within a warehouse, sometimes also demand forecasting. What we're not yet seeing as kind of a lot of end-to-end orchestration applications. That's probably the next wave. And that's also coming out of our survey actually, a lot of people say that the two methods that stand out as potentially generating the most value in the future is the generative AI and agentic AI and that's also the focus of the next survey that we're gonna run, identifying valuable use cases for these two technologies. But this shows that we are already on a way to go from simple optimization tasks that AI can support to actually end-to-end orchestration and that's partially driven at least by the growing expectations that we see from, for instance, omnichannel retailing. - Well, to add to what Matthias was saying, I think that adoption is becoming widespread because customer expectations keep rising. The reality is that same day and next day delivery leave almost no room for inefficiency. So when you are fulfilling thousands of time-sensitive orders, even small errors add up quickly and that directly affects, obviously, the service levels and the financial margins of companies. And moreover, AI is helping warehouses to operate with a level of precision that was never seen before and that omnichannel fulfillment requires. And this connects directly with what the Eva's research highlights. Omnichannel isn't just about adding more channels, it's about having full synchronization. And I think that AI is what enables that kind of end-to-end coordination and synchronization. So companies that don't adopt the level of intelligent optimization are simply less competitive. And in omnichannel environments, we all know that the speed is everything. So if you can respond quickly to demand shifts, promotions, or disruptions, it's clear that you will fall behind and afterwards, catching up becomes very, very difficult. - Sounds like the only downside to the technology becoming much more efficient and quicker is that the customer expectation is always rising. So as you said, leaving little room for error. And Eva, I know you've spoken to how customer expectations are rising rapidly in omnichannel as well, and your lab results indicate that AI is moving beyond isolated paid points to a systemic enabler across the operations. How does this omnichannel shift toward end-to-end transformation align with the specific AI applications Alejandro and Matthias are also seeing in the warehouse? - Yes. The findings of the omnichannel survey are very aligned with what Matthias and Alejandro just mentioned. AI, according to our findings, is becoming foundational. It's not optional anymore and it makes sense. Customer are asking for faster deliveries, but not only faster deliveries, more reliable deliveries, and this, from the operational side, requires much more precision. So in order to offer this seamless customer experience that omnichannel retailers promise, you need to have real-time inventory visibility. You need to have high inventory accuracy. So the shift we have observed this year is last year, companies were reporting the use of AI mainly on demand forecasting, mainly just to achieve a higher level of personalization from the customer experience. However, this year, clearly we observe an expansion in the scope of and role of applying AI just more on warehouse operations, for example, to orchestrate the fleet of robots, for example, to optimize the routes of these robots in the warehouse. Also, inventory management is a huge aspect in omnichannel because we need to allocate inventory through multiple channels which add much more complexities. This year, of course, customer experience and demand forecasting still is highly rated by our respondents, but as well is warehouse management, inventory management, transportation, and fulfillment. They rate these options, this implementation of AI in about 60%, each of them. So this shows the AI relevance is now more distributed across the operational supply chain backbone. We are moving from just AI in demand forecasting, just AI in warehouse to a more AI orchestrating the omnichannel supply chain. In this sense, it's still coming, it's not mature, but it's a trend that we are observing this year. - The state of omnichannel supply chain report finds that 80% of organizations report ongoing e-commerce growth and there's a 10% increase in those implementing omnichannel strategies. I think you've all spoken to the fact that AI is now a must-have to keep up with the rising complexity of fulfillment, but I wonder if you could speak to how much of a competitive advantage it is? - In your case, the growth of e-commerce and also the expansion of omnichannel strategies is bringing the need of not only faster, but also more precise decisions from supply chainers. So inventory position, as I mentioned before, is huge here because you have different channels, you need to provide this seamless customer experience and you need to allocate the inventory through different channels. But it's not only that it. It's where to prepare the online order, from where to fulfill the online order in order to satisfy these faster deliveries. This has implications in the network design, but also implications in the inventory management models and how to allocate these decisions, especially when you have large and huge data sets to handle. You also have real time variability in omnichannel. So all of these are aspects that bring a higher level of complexity that I think the old school tools are not as useful as it was in the past and is why I believe AI is becoming more a must to have than a nice to have just to handle this variability, real time, faster decision, more precise decisions. - Yep. People might ask, "Well, omnichannel has been around for a few years, we've somehow been able to do this so far, why do we suddenly need AI?" But I think to Eva's point, there're just so many sources of complexity. You mentioned a few, just think of, I don't know, omnichannel operations trying to grow the available assortment across channels means we have more SKUs, we have more fragmented orders, we have, as Eva mentioned, more fulfillment notes in a network that we somehow need to coordinate. We have, on the consumer side, for instance, things like tighter delivery time windows, in some cases. So at some point, both the existing software tools out there, but also sometimes the rather manual rules by which a lot of companies still operate are just not capable of managing that complexity anymore. At least not within the kind of timeframes that we have available. We could obviously run a traditional optimization model for three days and get the same solution, just that we need an answer in three minutes, not three days. And that's where AI really comes in. So to the question is AI a must-have? I would say let's say if your operations are really small, if you are a niche player and you may be more locally oriented, then maybe not, but in most other cases, if you're running a large scale omnichannel operation, there's just no way around AI. And it's also not just about having AI solutions as a shiny new object in your assortment, it's actually the difference between staying competitive or not. So it's no longer just something that is about saving a few dollars here and there. It's really the decisive factor of whether you're gonna be in business in the long run or not. - Yeah, in my opinion, I think that AI is a must-have, especially in the cases where complexity, precision, and speed are mandatory. And fulfillment today is far more complex than it was few years ago because now companies have every time more SKUs, more channels, higher inventory problems, and constant demand volatility. And in my opinion, the game is not about having the best AI or futuristic AI. I think that the game is about having intelligent optimization in place and in omnichannel environments, especially, if you can't adapt quickly and operate with precision, you are simply out of the game and you're not gonna stay competitive. And over time, in the end, your business will fall behind. - This episode is brought to you by the State of AI in Warehousing Report and the State of Omnichannel Supply Chain Report, which you can read at ctl.mit.edu/publications. Both reports, the State of AI in Warehousing Report and the State of Omnichannel Supply Chain Report, show that we're seeing notable increases in the perceived relevance of hardware like autonomous mobile robots, AMRs. How will this change the omnichannel and warehousing landscapes? - In the case of omnichannel survey, 63% of respondents consider autonomous mobile robots highly relevant, up to 50% last year. And it makes sense. E-commerce is bringing high volatility. Most of the retailers that operate in omnichannel, they are facing this high volatility piquing the demands, for example, Black Friday, Cyber Monday, the holidays. This means that they need to meet three times, sometimes even 10 times the regular demand. And autonomous mobile robot is bringing flexibility, flexibility to prepare these and fulfill these online orders. Not only flexibility, they are also easy to scale. And then the AI, on top of that, brings the intelligence that Alejandro was mentioned here. So it's helping to orchestrate this fleet of different robots. Sometimes it's a very complex because some of these robots interact with humans, are collaborative robots. So you need to orchestrate that in a human machine environment. So AI is bringing this intelligent on top of the flexibility and scalability that autonomous mobile robots are bringing. So that's why I believe practitioners are very much into this technology. - Yeah, there's a reason for why one of the core research areas that we tackled together with Mecalux in the first year of all our collaboration was and still is a design of better algorithms for AMRs because the increased importance of this technology in warehousing is an immediate reflection of the increased volatility in the marketplace, especially in omnichannel applications that basically conflict with labor constraints. And so basically allow companies, for instance, to quickly scale to respond to demand volatility. Now the question is why do we need AMRs for that? Simply speaking, because AMRs can adapt to, for instance, changing demand patterns. Conveyors typically can't. So a traditional conveyor system is put in place for the long run and needs to be designed exactly to capacity. An AMR system, if you are seeing a, I don't know, 5x, 10x increase in demand for a couple of days, you can just increase the number of AMRs that you're operating for that period of time. You could add additional agents to the system and that automatically scales. While you cannot simply do this with more fixed infrastructure investments like, for instance, conveyors. So the one thing that I would like to point out though is that it's not just about the hardware, the admittedly impressive hardware, it's always cool to see a robot move around a warehouse seemingly on its own. But if you only focus on the hardware, you basically have a bunch of moving robots, but you don't yet have an optimized system. So part of the reason why we did the research with Mecalux is to make those robots more intelligent, to actually coordinate across multiple robots, to really make them behave more human-like, more intelligent instead of just following fixed predefined rules that some engineer once upon a time made up. And that's still kind of a challenge. And that's where, again, AI comes into the picture because obviously these robots use relatively well-established technology like object detection, image recognition, to move around without running into anything, but to really make them smart, to really make them take their own decisions as to which task to take on when and how to distribute the workload across a fleet of AMRs. That's where real state-of-the-art AI models are needed. And that's basically what we've been focusing over the last year or so. So maybe the future of warehouse operations isn't just about automation, it's actually about adaptive systems that can scale to demand, but it can also coordinate dynamically across multiple technologies, not just AMRs. - Yeah, and to build on top of that, because it's also the upfront investment that companies need to do. Matthias, you also mentioned that, for example, the automated storage and retrieval system are much more rigid, but also the investment upfront is higher than the investment you need to do with this flexible model, especially if you combine with a robot as a service model, for example, that is really flexible. So yeah, I think also this upfront investment is another incentive for companies to invest in this. - Which is interesting because you actually see completely new business models emerge where- - Exactly. - I mean, as you said before, previously a company had to decide whether or not to spend a very large amount of money to build a very fixed asset. Now there are companies just renting out robots, simply speaking, which also makes it more affordable to smaller players who might have previously not been able to pull off a fully automated warehouse. - Exactly. - So the barrier to entry is lower? - Much lower, yeah. Especially with the robot as a service models. - Yeah, because typically for, you know, for automated warehouses, we were talking about investments that we're talking about millions usually. But when it comes to AMRs, I think that with a much more affordable budgets, you can have a fleet of AMRs running in your facility. So this is why another main advantage is that the affordability of this kind of technology is way more attractive for customers now. - Speaking to the human element, there's a persistent fear that AI will cause widespread job loss, but data from the State of AI in Warehousing report actually shows that AI and automation are expanding the workforce instead of shrinking it. Why do you think that's happening and what kinds of new roles are emerging? - Yeah, so we were not necessarily surprised by this finding, but we didn't think it could would come out this clearly, to be honest, from the responses that we got. But on a very high level, what's actually happening here, first of all, warehouses face rising e-commerce volumes, generally rising demand, and at the same time, higher service level expectations from their customers. But at the same time, at least in most, let's say, industrialized markets, we are facing actually a labor shortage. So we are actually struggling to find skilled workers to run a warehouse. And that's I think where AI comes in and fills the gaps and helps us improve throughput so that we can handle the increased demand and the increased requirements by our customers with the same or only a slightly increased workforce. So that growth in volume and the growth in actually workload probably offsets the undeniable efficiency gains that you would get out of automation. So the net effect is positive, not negative. So that's why if we look at our survey responses, most of our respondents actually say that our workforce size increased with the adoption of AI and yet we're not actually seeing fewer jobs, but different jobs. So we see, obviously, a couple of highly repetitive tasks being fully automated away, but at the same time, we see new roles emerging. We see on the high end of highly skilled labor, obviously, a bunch of AI and ML engineers that are needed. We need some automation specialists, data scientists and the like. But at the same time, at the level of the people who are actually working on the shop floor in the warehouse, we need more skilled people who are able to basically control the more automated technology that is supporting them in running the warehouse. So simply speaking, there is a structural shift. The human role in the warehouse changes from doing the entire job on your own to basically supervising automated systems, handling exceptions, helping the system optimize, helping improve the design of the system. So those are the tasks where human intelligence actually still adds the biggest value. And that also might be a factor in things like workforce retention, which again, if we look at the responses that we get from our survey, we see not just a general increase in workforce due to the adoption of AI, but also an increase in kind of worker satisfaction, job satisfaction, and therefore, potentially retention. So in a nutshell, we are not necessarily seeing humans being entirely replaced by AI, but we rather see them kind of move up the value chain. But to enable that, we obviously need to upskill our workforce. So basically help the people who are currently doing the repetitive tasks to stay relevant by being able to supervise the systems that take that away from them. - Eva, I don't wanna let that go by without saying that in your other work at CTL, you work a lot with upskilling the workforce. I'm wondering if you could speak to that need that you've been seeing? - Yeah, that's a great point. Yeah, this is my role as the director of online education at the center. One of the things and trends we have observed recently is more and more companies are coming to us with the need of upskilling their workforce. And it makes sense because one of the key things we have learned through these years is that AI tools are very powerful. However, you need to understand how the tools work and have this foundation. So online education is very powerful. In order to understand how the models work, you need to be able to challenge the output and adapt the output to your specific context. - And Eva, the omnichannel report highlights the cost to serve and returns have risen significantly, which you mentioned earlier. How's AI helping companies move from just building capability to actually managing margin at scale? - Yeah, you mentioned returns. Returns is a big challenge in omnichannel because online returns are about 20% higher than traditional returns. In certain industries like fashion, return rates can be 35 to 40%. So this year, we have observed companies piloting AI tools to improve the online customer experience. I mean, for example, these virtual try-ons, they are incorporating AI tools on the virtual acts in order to provide much more information to the customer. Same thing with they are incorporating sizing and fit prediction models. All of these tools are improving the online customer experience, provide much more information, and at the end of the day, are reducing the return rate. So this is one of the areas. Regarding the cost to serve is rising, I'm not surprised about that because we also already mentioned the fragmented orders, the faster deliveries promise, the higher return volumes that I just mentioned. So AI is trying to improve this cost through a more accurate demand forecasting, through the optimization inventory position, minimizing also split shipments or yes, enabling more dynamic routes. So the shift in focus that we have observed this year is companies were investing in the last years more in visibility, in automation, in the actual infrastructure. However, this year, they are investing more, and our capabilities that they are more material and this year, they are investing more on these AI tools that try to enhance, optimize this network performance from the end-to-end supply chain. - And based on the warehousing report findings, where are companies actually seeing the fastest returns on AI and automation? Which is another way of saying where should companies perhaps invest first or continue to invest? - So from an investment and return perspective, companies are prioritizing to adopt AI where the return is clear and measurable. So we all know that obviously adoption really accelerates when the impact can be tied directly to financial performance. And the strongest use cases tend to be in areas that affect the cost per order. Things like improving the inventory accuracy, optimizing the picking workflows, or allocating labor to tasks in a more efficient way. And I think that companies aren't just applying AI everywhere for the sake of it. They're being very pragmatic in this sense and they start with the areas that can quickly improve key financial results and they can achieve the first quick wins. And once they see that there is a clear and measurable impact, then it's the time when they decide to expand it to the rest of the operation. - Yeah, currently the strongest ROI is probably being observed in execution layer optimization. So basically singling out specific problems within a supply chain, within logistic systems that we can optimize better through the use of AI. And so here the traditional metrics are things like cost per order and service level. If we look at the long term, so where do we really need kind of longer term investment into capabilities? Like, what types of outcomes do we wanna see? I think it's all about orchestration. So the longer term plays that we won't be able to see a short term ROI kind of immediately after implementation are things like end-to-end orchestration of fulfillment systems or fully autonomous decision making. If we ever wanna get there, if we ever want to have a fully autonomous supply chain that basically makes decisions on all levels more or less independently from any human input, that's a very long term play. We're far from that today, but there are companies investing into this for good reasons, but that's an area where we won't see a positive ROI on these types of investments in the next two, three, four years. Same thing I would say with digital twins, actually, we're currently working with Mecalux on a digital twin of an advanced automated warehousing system. Those are long-term investments. These are capabilities that are hard to build, but that will pay off in the long run. And for all of these longer term plays, what's really required is solid data integration. That's currently one of the main bottlenecks, to be honest, that we see for larger scale deployment of AI is the availability and also ingestion of high quality data and integration of these tools into existing IT kind of environments. Then to Eva's point about upskilling, we need stronger internal capabilities. We cannot solely rely on vendors and consultants to do this. We need some people on the ground in the company who know how these systems work. Scaling more advanced, longer term AI deployments really takes investment of the entire organization, not just building fancy algorithms. So the short term ROI that we currently see mostly comes from optimizing the individual tasks. The long-term value of AI really comes from end-to-end orchestration across tasks within the supply chain. - And can you all tell me about a challenge you faced when doing this research and how you dealt with it? 'Cause we see the findings, but we don't get to hear the stories behind the scenes of what you all encountered. - Yes, in the many challenges, but one to highlight this year, as I mentioned before, we wanted to deeper into the role of AI omnichannel and we want to shift from are you using AI to which specific AI tools are you using and where? So to capture this detail answer, it was a challenge because we need to keep the survey less than five minutes. So the approach we follow is to provide a curate list of AI tools, those that we have already identified that are the most common tools that companies are implementing. We map those tools with the key functions in omnichannel was customer experience, warehouse management, fulfillment management, transportation management. And then it works. Finally, we have more than 600 practitioners completing the question and we identify a clear adoption patterns. Was interesting to see that, for example, large language model chats and gen AI copilots dominate customer experience. However, random forest and deep neural networks lead demand forecasting. So the insight here was that AI adoption derived by function is increasingly embedded across omnichannel and supply chains. - That's an important point 'cause people may not think about the fact that you wanna get a large enough sample size. So you wanna make the survey short enough that you can get a good size, but also you want as detailed responses as possible. So you have to strike the balance. - Yeah, I guess that's gonna be a challenge that we'll also be facing this year. Actually, you know, a follow up survey because this year, we're actually gonna do a follow on to last year's survey focusing on specific application specific use cases for gen AI and agentic AI in warehousing. And it's gonna be hard to really get to the kind of root of what the value driver is by just asking a bunch of multiple choice questions. So we will have to engage our respondents more deeply. We will have to ask them to provide more context, more detail, which is something that's hard to do on a pure survey based method. So I guess we will actually combine a survey with more structured interviews of selected respondents just to get a little bit more context around the sheer statistical observations that we're making in the survey responses. - So everyone look out for that survey when it launches later on. Looking ahead in terms of key recommendations for companies for Mecalux and ILS, given the trends that you found, what should business leaders do in the next 12 months to make real progress on AI in their warehouse operations? - What I would recommend is, first of all, pick a couple of proven wins when you think about at least short term AI deployment. Start with use cases that have proven to have a high ROI, things like inventory optimization picking, through productive optimization and the like, and focus on areas where that impact is actually measurable. Because in order to convince your board or whoever is the eventual decision maker for scaling, you need to show that whatever pilot you just ran is actually successful. Then the second recommendation is define clear KPIs for such a deployment early on, because if you can't measure it, you can't repeat it and you cannot convince your decision makers to scale it. So those KPIs could be things like cost per order, service level improvements, inventory trends, that kind of thing. And lastly, create small and cross-functional teams to actually run these deployments. So you wanna have people from operations in there that actually understand the business problem. You wanna have people from your IT and data department who understands the ecosystem within which any future AI tool needs to live in. And then you obviously need the folks who know AI and machine learning really well to design the right solution for the right problem. AI should never live in a silo. It should not just be another department within the company that does its own thing. Because if we look at how some of the deployments fail, and most of the time, it's not a technical problem that makes them fail, it's an organizational one. - I think that the first step to be successful, I think that it's to be pragmatic. So don't start with how do I use AI or where do I use AI? Because I mean, this is not accurate enough. I think that companies, what they have to do is to start with thinking, first of all, where is my biggest operational bottleneck, or simply speaking, where do you think that my operations are losing money nowadays? And once this is clear, then try to identify and focus on one or two high impact areas, which could be things like forecasting accuracy, labor planning, or slotting of optimization, whatever fits best for any company. And it's very important that you focus on areas where the results can be measured clearly and also achieved in a relatively short period of time. Second, AI is only as good as the data behind it. So visibility, accuracy and system integration are critical. And finally, I would say that think pilot first and scale second. So run control pilots, measure always return on investment, and then once all of this works, expand it. - So my suggestion for business leaders is first start with the problem, not the technology. A strategy first, I mean define your service promise, which channels do you want to offer? Which service level? How fast you want to deliver them? Design the network appropriately for that environment and then enable and decide which technology, which level of automation, which level of AI tools you want to implement there. I think in omnichannel, inventory precision is non-negotiable. You need to provide real-time visibility and high inventory accuracy. And last but not least, upscale the workforce. I truly believe that the ability to use critically and strategically AI tools is going to be the way to go. Talent and capability is I think what companies need to get the most of technology and AI solutions. - That's a great point to start with the problem rather than saying, you know, what's out there, what can we use? You know, how can we use it? To close us out, for ILS and Mecalux, one of your findings was that there are two prominent valuable methods being used in warehouses, generative AI and agentic AI. You spoke to this earlier, but you're gonna focus on agentic AI in your next report. I'm curious if there's anything you wanna share about that focus. - Well, gen AI and agentic AI are currently, for good reasons, the two big buzzwords in the industry. Even though we don't see the bulk of the current AI deployments to actually fall into this category. Yet companies see these two methods or families of methods to be the most promising in terms of generating long-term value because let's say the AI deployments that we're seeing today are optimizing individual tasks. Those are gonna become pretty commonplace kind of a commodity pretty soon. So those are no longer gonna be real competitive differentiators in the long run. But the capabilities that we may be able to develop using gen AI methods and agentic AI methods go beyond that. Those are really kind of focused on system-level orchestration and that's where we wanna dig deeper. There are a couple of companies out there already who have piloted or potentially even gone beyond pilots with gen AI, but also agentic AI applications. A lot of them are still rather nascent. There are a couple of startups in this space, for sure. But in this new survey that we're gonna launch this year, we really wanna understand what are the concrete use cases that people see. Because when I speak on a very high level about system level integration and orchestration, what does that really mean? Like, which decisions across the supply chain do practitioners perceive to be the most important to orchestrate? How do we really enable the people who are doing this today in a rather manual or kind of traditional optimization based way, how do we enable them to make better decisions using these more advanced AI tools and what does it take to actually deploy them at scale? - Yes, while generative AI is great for things like documentation, training, and user support, agentic AI is where we really see the biggest operational impact. And agentic AI, in the end, can make decisions, take actions, and continuously optimize processes in real time. So we think that it's gonna be very important for us to understand how this can be applied, you know, in our warehouse environment. Because in the end, this means that, you know, applying this kind of technology means being able to dynamically adjusting workflows, prioritizing tasks, and responding to disruption without constant human intervention, which is what is happening nowadays. - So look out for that launching later this year. That wraps up this episode of "Supply Chain Frontiers". A big thank you to Matthias Winkenbach, Alejandro Gonzalez, and Eva Ponce for sharing their expertise and insights into the current state of AI adoption in warehousing e-commerce among other findings. "Supply Chain Frontiers" is recorded on the MIT campus in Cambridge, Massachusetts. Our sound editors are Dave Lishansky and Danielle Simpson at David Benjamin Sound. And our audio engineer today is Kurt Schneider of MIT Audio Visual Services. Our producers, myself, Mackenzie Berry. Be sure to check out previous episodes of "Supply Chain Frontiers" at ctl.mit.edu/podcast or search for us on your preferred podcast platform. I'm Mackenzie Berry, thanks for listening and we'll catch you next time on "Supply Chain Frontiers".