Name: Supply Chain Planning Reimagined: Embedded AI that senses, explains, and optimizes
Uploaded: 2026-02-18T17:16:14.370Z
Duration: 57 min 24 s
Description: Supply Chain Planning Reimagined: Embedded AI that senses, explains, and optimizes

Transcript for "Supply Chain Planning Reimagined: Embedded AI that senses, explains, and optimizes": Okay. Here we are. Can you hear me, Todd? Yes. I can hear you. we are on stage. K, guys. So sorry for this, but we are facing some technical issues, and we still have to get our moderator back in. But in the meantime, welcome to the webinar, and I'm happy to see the discussion is already ongoing in the chat. That's great. That's great to hear. So we're still waiting a couple of minutes for our moderator. So you're supposed to be here to answering the questions. Let's see. No moderator, no party. Correct. That is very true. Yeah. I don't know if we should maybe just start with the introduction, or do we wait for Ella for that? I I think we can introduce ourselves. Right? And then I would say so. yes. And then I think we have the first the first question. So, Laura, why don't you go ahead and introduce yourself first? K. So hello, everybody. I'm Laura Tozzo from SAP. Well, that's hopefully clear. But specifically, I'm in the product management for SAP integrated business planning. I've been with SAP for seven years and always in the planning area. And in the last couple of years, my focus has been mostly on artificial intelligence, and, of course, then nowadays much more on generative AI and agents. Sure. And I'm Todd Stenger, also from SAP. I work in product marketing and cover, topics for supply chain planning around demand planning and, and AI, topics more generally. I've been in supply chain planning for over thirty years now, mostly at software companies. Right? So I I think it's a lot of I've seen a lot of things over the years, and I'm very excited about where we're going with with this idea of AI and what it can potentially do and and some of what it can't do. Alright. So, I thought that we would start with, the first question that they had. I'll I'll go ahead and read it off, which is, since Ella is not here, organizations have been trying to fix resilience for years. However, given that change is the only constant, how do we move beyond simply being reactive to building a supply chain that can actually thrive within a volatile market landscape? So I guess I'll I'll start with the, with some comments, and then I'll let Laura chime in. But, I think that as supply chain leaders navigate these uncertain markets, it's fundamentally that they need to make informed end to end plans and decisions across the entire value chain. And I think historically, both in midterm and long term, it's been a lot about planning within the four walls of their company or within the supply chain that they really control. And that, you know, we really need to move towards considering both upstream and downstream impacts of disruptions as well as disruptions both upstream and downstream that are going to impact the supply chain. And that allows you to get more early warning and to basically have a better understanding of the entirety of the supply chain. And really, in order to do that, you have to have a robust planning platform that allows those leaders and planners to make data driven decisions and act swiftly and stay ahead of dynamic markets. So it's kind of like a general overall. And I think Laura has got some thoughts on it just from a more technology perspective. Yeah. I mean, the the the thing is, I think, the idea is to make making sure that you react fast to disruptions and to, yeah, whatever can happen as fast as you can. Right? I think even AI is not going to be able to break everything that can happen, but the goal is to make it possible in, yeah, in the fastest possible way. And here we have our moderator back. Hello. I'm so sorry. I had some terrible Internet issues, but I'm here now. But I'm so glad that you guys got going about us. So I believe we're on question one. So. let don't let me stop your flow, Laura. I I was just commenting on Todd's answers, but I think we are good to go and and proceed with the next questions. Brilliant. So let's look more specifically then at how SAP approaches, how how SAP's approach can specifically enable the company to strengthen its planning resilience without sacrificing efficiency. Yeah. I think I think if you look at sort of this idea of an AI plat you know, powered platform, you know, streamlining planning and decision making, you know, the there's a couple different pieces of this, right, is that, you know, that that one is, you know, connecting to an extensive business network and enhancing collaboration with partners. I I talked a little bit about, you know, it's gotta be beyond the four walls to really be more resilient and and getting, you know, partner suppliers, customers. And then, you know, the idea of integrating the execution applications is really big. And, really, we're we're trying to go for if you kinda think of sort of pillars of of of big things from our approach is that, you know, a, we want to make sure that you have a lot more autonomous planning. And then the other piece is, really, once you've done planning, understanding what has happened, and you might call this planning decision intelligence, which is, you know, how did we do as we were planning? I think a lot of companies, they create a plan. They might be doing reaction, firefighting, all of these topics, but they never really go back and say, was there something we could have done better? And where did we make decisions that are different? Right? And so I think that that's huge. And, you know, as part of this, you also have to have, you know, kind of decision centric scenario simulations. You certainly have to have, you know, better user interfaces. And, you know, as you look at from an SAP perspective, we're really doing a lot of things where, you know, we are moving away from necessarily, you know, custom, screens and UIs to, you know, more of a chat GPT experience through our dual assistant. Right? So that a lot of things will be done through that. And I think that in and of itself is fundamentally AI. Laura, did you want to expand on that? Well, I I think, as Todd said, right, it's in the end, it's about the interaction of the users with the system that's changing a lot. And indeed, at SAP, here, we are focusing much more now on the user experience and how natural language is going to change this in the future. So let us look now at the AI aspect of of 2026. It wouldn't be a it wouldn't be a webinar in '26 if we didn't mention AI, And we're at the innovation frontier, really. So we hear a lot about AI. There's so much hype around AI, but embedded AI is another step beyond that. And it's no longer just about getting systems to talk to one another and automate processes for us. It's about allowing AI to sense what is expected, which is obviously really key in supply chain. So what are the key pieces of information that allow this embedded AI to make informed decisions? If I throw it to you first, Laura. Well, I mean, first, I want to to say that the the the central role of data hasn't changed. Right? So everybody who runs the supply chain knows that data is the foundation of everything, and it has been the foundation in the past also for, you know, like, traditional AI, machine learning for forecasting, and these kind of things, and that remains true. Right? You still need to have a a good basis of data to make everything possible. But now when we talk about the future and more around agents that are embedded and sense things and and so on, what becomes really important is the data around processes. Because while, of course, some pieces of these can be generalized, some problems are things that happen for multiple customers, at the end of the day, how the customers operate their supply chain, the processes are always specific to them. And so to make sure that the agents are really able to do their job at the best, the process information, the process data starts to become really a key differentiator compared to before. And in the conversations I've had with customers, this is the piece that is still not really there very often. The information about the process data is in the hands of people, of the users, of the planners, but it's not written down. You cannot tell an agent how to do their job unless you can write down what they are supposed to do. So that's going to become really key in the future. It's really just about making sure that the agent is speaking in your language, understanding your needs. It's so will is there a lot of, like, prompt engineering there from a from a user point of view? not prompt engineering. It's context engineering a lot, so that's a little bit the shift. But because it's not about how you phrase the request, but really how much context the agent has to understand the request, to understand which what are your data, what are your products called, what are your locations called. And then it's about writing it clearly enough that the agent can follow through on the instructions. So those are the two pieces. Fabulous. So when it when it comes to that embedded AI, it's obviously so easy to just allow it to kind of take a take an idea and run with it with very little human interaction, which is why it's so important that we trust our agents. So why is it important now that the AI explains its logic back to the end user? Well, I think the the trust is the main issue here. I mean, it's been also in the past with more traditional AI solutions. We have seen already that users already have problems trusting the results. If it's too complicated in the back end and it's a black box, why is this result as it is? Right? So what did the system do? So explainability was already super important, but it becomes now even more important because in the past, although they were AI black boxes, they were at least deterministic. Like, you could give it the same input twice, and it would give you always the same output. Now this is not true anymore for agents. We have the probabilistic component coming in, And so you really need to be able to explain to the user what are the steps the agent is going through. Because otherwise, they will never be able to trust the decisions, the suggestions that the agent is giving without being able to understand the steps it took. I think it's really important. I think explaining like, explainable AI is is such a massive thing right now because it is so easy just to let it just take it away, and and we don't have we can sit back and not even really interact, but it's quite dangerous to business. Well, Definitely. I it it's interesting when we talk about agents. And and, again, Laura kinda touched on this is that the other thing with agents is, you know, putting the guardrails around it. Right? So if you're gonna have an agent that allows it to rebalance stock. Right? So, you know, I I it turns out I don't have enough stock in a DC, and I can go get it from another DC. Right? We know that historically, through optimization techniques, people a lot of times didn't wanna set that up because optimizers would just start, you know, sending stock back and forth, and it got really hard to deal with. But from an agent perspective, if you're going to let it do that, you also need to tell it when. Right? And and when can it do that, and how much can it do it, and when does it need to ask the human. Right? So you might and and, you know, this is not new to the business world. Right? So a lot of business processes have these guardrails that say, look. A planner can can move so much or can approve so much, and then then it has to go to then it has to go to management. Right? And so you still have that concept with agents that you have to deal with. And and so, you know, the agent goes through, finds five problems. Four of them are one. You told it, look. You're just sending 20 units, you know, somewhere. Totally fine. Go ahead and do that. And but the, you know, the the last one is it's gonna have to move up, like, 10,000 units. And and then you start looking at this thing, how do I constrain what it's doing? So it's not just explaining it. Also, how do I make sure I've set it up so that it follows the business rules that I have, and how can you do that easily? That's kinda what Laura's talking about, the context engineering. Yes. Making sure it knows its boundaries, really. Making sure that it knows it can be responsible for 50 items, but not 500,000 items. Right. So. when we look at that at the technology stack then, how are we how are the emerging tools different from the tradition like, the traditional automation that we've seen in the past? Well, I think the the main change now is how this is becoming much more easy to use. Right? So it used to be that AI means collecting a huge amount of data, cleansing the data, and, you know, having a data science scientist or maybe a team of data scientists in many cases working on fine tuning the the the models and making sure they speed out the the best numbers. And don't get me wrong. There is still a place for that. Right? That's still applicable to many things because GenAI and agents are not necessarily good at math. So there's still a place for that. But when we are talking about agents, engine AI, and the future of AI, it becomes much easier to just define an agent and to instruct an agent. You don't need to have any technical knowledge, you don't need to be a data scientist, you can just write down instructions, and then the agent will just, yeah, sort them out and do what you're asking it to. So it's making it much easier, the lowering the barrier for usage of AI, which, of course, brings, of you know, also some trouble with it, because then you have to make sure that the people are also trained, that they understand what's possible, what's not possible, how to put the guidelines in place, as Todd was saying. But in general, I believe that the, you know, the new tools will make it once we get the foundation in place, which is where we are now, and it's going to still take a second, once we get the foundational pieces, like APIs and so on in place, it's going to scale much faster and be much easier for everybody to customize things as they want. So when we look specifically at the SAP solutions then and the broader SAP tools, what what can apply to the supply chain planning process? You're gonna say everything. Well, yes. But, I mean, in in SAP, of course I mean, in in I b p specifically, we have, of course, the existing tools, tools, AI AI tools, like the demand forecasting. They use still the traditional method and most likely still need, partially, data scientists to be scientists to be on board to to fine tune. But we are also in we have also introduced last year quite some use cases around explainability that are just very simple to use. You know, it's a one click thing, and then you just get your answer. You don't need to do anything special. And now the direction we are going, Todd already mentioned Joule before, and we have also a whole local, no code approach here with the Drule Studio, for example, where also customers can define their own agents and their own skills. And this is what I meant with it's going to become easier than also for everybody to just define something that that's specific to their use case. Yeah. I I I think there's another another piece that's a little more foundational from an SAP perspective too is that, you know, that that people, when they see a chatbot box, you know, the the assumption is I can ask in anything. And, you know, general purpose LLMs aren't necessarily trained on your company's data. And so if I ask it a question, you know, if I ask it a question like, show me open orders. You know, the definition of open orders is different across all kinds of companies. Right? Like, the what does an open order? Right? Does it mean one that hasn't been fulfilled completely? Is it one that has been partially fulfilled that's still open? Is it one in the future, one in the back, one in the history one in history that hasn't been fulfilled? That's not open. We call that backlog. All this terminology, and CP perspective, we have something that we call document grounding, I think, is a fairly standard term. But that allows customers to basically, you know, put all the documents relevant to their implementation, their company, their supply chain into, basically, a container that then we use. So when you ask Jewel a question, like, show me open orders, it can understand what open orders means to your company. And kinda to your point about, like, the prompt engineering and and everything is that do you need to know that? In a traditional LLM, if you just go out to chat GPT or something like that and you wanna ask it a question, you have to give it a whole bunch of information around that. You have those specific prompts. But by using document grounding, we allow, know, end users in SAP systems to use the terminology that they want to use for your supply chain and get answers that are relevant to them. And that all ties in with some of other tools that Laura talked about. Yeah. So it's really important, obviously, the whole way through, isn't it, just to get to make it understand your needs, your company's needs, how your language choices. Yes. Just completely shaping it like it's your child, essentially. Mhmm. I mean, I think it's always a good metaphor to think about agents as the kind of these new hires that. come fresh from university. They have no idea about the company. They have no idea about the processes. You just have to tell them, here, these are the tools that you can use, and these tools does this, and this tool does that. And this is the order in which you should do things, and, you know, this is your scope, and this is what you're supposed to do. So always think about agents in that sense. You still need to give them all the context, all the information they need to carry out their tasks. And they also have the very keen optimism of a new hire. 24 a day. Right. Exactly. No days off. So, obviously, we've spoken a bit about people and, like, people's role in in in AI now. But at the end of the day, a system is only as good as the person that's operating it. So how does reimagining the planning process actually change the daily life of the supply chain planner of the future? If I throw it to you first, Todd, on this one. Sure. I I mean, you know, I think it's obvious to everyone that that the tasks the planners are doing are going to change. Right? I think if you really think about, you know, planners day in and day out, I I think when I've talked with customers and, you know, companies and and I look at what what do people do all day, you know, planners do, and I'd say a lot of what they do is investigating problems. Right? So they're investigating. They're manually adjusting plans. They're working with spreadsheets maybe because, you know, they need additional information. You know, they're they're making phone calls. There's a lot of there's a lot of pieces that I think that we can start start to automate. Right? And so that's you know, that as we look at AI and agents is that, you know, much of the time that that planners are spending today, we should be able to automate a lot of that. Right? So the investigation is a very good example, you know, that you might have an an agent that, you know, that recognizes there's a disruption, creates a couple scenarios for you, you know, runs the evaluation, and then comes back. And and the planner now is saying, you know, okay. I can quickly evaluate the scenarios that you gave me. I'm not creating it. I'm not adjusting models. I'm not deleting data. I'm not adding data. All of those tasks they may have been doing. And now they're really just making an informed decision. Right? And so that I like to think of it a little bit as moving away from directly fighting fires to, you know, deploying the resources to fight the fires. Right? So that we're not, you know, we're not saying, oh, I have to go in. I have to get the hose out. I gotta, you know, spray. I gotta decide where to spray it. Now they're saying, you know, okay. There's a bunch of fires. There's already people on-site, or there's another one that needs people on-site. And so I'm doing more of that coordination and orchestration, as opposed digging into all the data and and, you know, playing the system. So essentially moving from, you know, reactive to proactive. Laura, Yeah. I feel like I interrupted you there. And I I think also something that we've been hearing from customers is that there is an expectation, and, mean, I we'll see if this will become true. But I think it makes sense that once you take away, as Todd said, from the firefighting of the little fires every day, which causes them people to be very silent, right, and con concentrate on a very specific piece of the supply chain because they only have the time and the bandwidth to focus on the problems that arise on this specific piece. Once you take that the the problem of the mass of firefighting away, then the expectation is that planners might evolve to from from, you know, very specific planners, like, demand planners, supply planners, maybe sometimes, you know, regional planners, local planners, that they can maybe even evolve to a different type of role in which they are kind of end to end supply chain planners and have this more on a holistic view on on some product groups, for example, and therefore also be able, as you say, to be more proactive and take better decision overall for the throughout the entire supply chain. So, obviously, we're talking a lot about, like, freeing up people's time here. So in terms of agility, how does an AI enhanced environment allow a team to stay just more focused when the supply chain is under pressure? Well, I mean, at the end of the day, it's what what there is a a new buzzword out there. It's been out for for a bit, actually, which is decision intelligence now. Right? And I think that at the end of the day, that's the goal, that we want to make sure that, as we said, that the firefighting stops, but more than anything, that AI is supporting planners in taking decisions. So it should take away all the tasks tasks that are, you know, taken away from the big picture from the how do I take the best decision and really enabling them to focus on what's important, being able to, you know, for example, compare scenarios easily based on very specific APIs. And, you know, having an agent or AI pieces of AI in the back end, just doing the analysis for them, and really just let the human in the loop be the one taking the decision. And so at the end of the day, that's where, you know, not only the efficiency in terms of less hours spent in firefighting, but the efficient in terms of making better decisions overall is going to come from. Right? That that's the, I think, the overall goal for the supply chain. So you're talking oh, sorry, Todd. Well, I was gonna I was gonna add that if you if you think about you know, like, I talked a lot about what planners are doing day in day to day. You know, there's also if you if you go back to the business process of the supply chain is that, you know, again, there's there's a lot of decisions that need to be made that are impacting the business. And, you know, if you think of just a sort of a scenario where, you know, there's a a disruption going on, you know, maybe maybe, you know, this US Canal gets blocked. And if you think of kinda how that played out, right, it was, you kind of all hands on deck. And and, you know, management is asking asking questions. Right? Like, what questions are they asking? They're asking, you know, can we reroute things? Can we you know, do we have stock we can bring over land? Do we, you know, do we reroute ships? Do we you know, there's a lot of questions that are there. And and if you think of the amount of effort the planners had to go through to answer those questions, you know, if you take the, you know, the AI enhanced environment and if you can do that quick quickly, hopefully, you know, as we get into the agent world, the agents have already queued up several scenarios for you. But even without that, you know, if if you if a if a planner can just tell the system, hey. You know, go zero out the the routes through the Suez Canal and replan and then do that quickly. Now they can actually, you know, bubble that information up to upper management that says, here's what we're gonna have to do. Here's what it's gonna cost. And and, you know, here's how we get out of this situation. You know, those those types of situations are happening more and more and are happening, you know, in in you know, as all disruptions are, they happen whenever. And. so that this idea that says, I don't have to go off and spend two days generating a scenario or, you know, rearranging my model in my supply chain to understand what the options are is mass. And then the other piece, as you talked about explainability, is being able to coherently explain to upper management, you know, what those financial trade offs are and where those numbers came from. Because it's not just about the planner understanding it. It's it's whoever ultimately is gonna make a business decision that needs to understand that as well. And as the world becomes more volatile, we're gonna rely more and more on agents like that that, you know, can can stop the disruptions before they are a problem for us. So you're Laura, you mentioned, you know, we're giving the we're giving the humans a chance to take the important decisions, but what are those important decisions in this new in this new workflow? Well, I ideally, it's, you know, it's not going to be only I mean, nowadays, we hear a lot of planners are busy in taking the decision on whether to move, you know, product a from location a to location b, so that this one a customer order that's, you know, happening tomorrow, it's going to be fulfilled. But in the scheme of things, is that this customer order, really the most important order you need to fulfill, you just don't know that. When you're in firefighting mode, you're just looking in the next twenty four hours and try just trying to make sure that that's going to be the best it can be. Right? But if you have an agent already looking forward for you, then you can also can imagine coming into in in the morning and having already a list of, oh, look, you are going to have some unfulfilled demand for this and this and that customer, and the reason is you're you're not producing enough, you're not getting enough, whatever it is. And you could already have the, you know, the sorted list, for example. And then you could maybe decide, well, you know, this customer order, actually, I don't need to move product a from location a to product to location b, but it would be actually better in the overall scheme of things to think about increasing production this week so that the next week, I can satisfy instead, you know, a bunch of other things that are coming from more important customers, for example. And so I think right now, the the firefighting is just the visibility is not there, and overall, you know, the the important decisions should be the ones that have a return on investment in hand for the company. Right? And I think many of the things that planners do today are not necessarily needed, but they are just the way it's done. You know what I mean? it's yeah. It's really important, isn't it, just to not have to just look a day ahead. You can be more strategic then and look couple months ahead or maybe even a year ahead Yes. and plan for the future instead of just blinded by the flames all the time. I I think there's also I think there's also what Laura talked about is this idea of a broader spectrum or a broader viewpoint. If you think of a lot of planners today, a planner might be responsible for 50 materials. Right? And and I think one classic, example that kinda ties in with our this idea of explainability and and what you actually look at and how you make decisions, is that a lot of times, we've had optimization for years. And a lot of times, the complaint I would hear, and I I always thought this was really funny, is that a planner will complain that the optimizer said that predicted they were gonna stock out in six months. And a planner who has 20 products says, I would never do that. Like, this is dumb. Why is the system doing this? I would never do that. But it gets to that bigger context that Laura's getting to is that, you know, it could be that stocking out one item will will mean that you'll fulfill eight different customers' orders. Right? And one customer will be impacted, but they don't have the scope. And when you think of that, they need to understand that. They need to understand the the impact outside of their world. When you start talking about agents, the agent can already go look at that. And then if it comes up in the morning, it says, you know, if you think agentically that you say in the morning, it comes in and says, hey. Your your products are gonna stock out in six months. Here's why. Here's the other things that are going to be impacted, you know, so that the planner doesn't sit there and go, look. I'm gonna get dinged on my KPIs because, you know, because the optimizer didn't know what it was doing. Now they look at it and say, oh. And then there might be a decision that they could take that's outside of the scope of what you let the agent do that says, oh, we have another supplier that could supply the bottleneck item for this. It's just that it costs us a lot of money, and we aren't gonna let the system do that automatically. And then the planner's now making an ROI decision that Laura was talking about. Says says, you know what? It is worth it to us to do that. One company that's going to be impacted for this instead of eight is also a company that turns out that we've got you know, we've had a lot of short orders for, and we're having a bad relationship with them. And, you know, there's some broader context that that you're not gonna let the agent do. And now there's okay. It makes sense to spend that money because it's about the relationship with the Right? And that's the type of decisions that you want planners to be making and not, you know, the worrying about the fact that it's stocked up in six months and not have any idea why. So. Well, thank you both so much. I would have said at the very beginning that you don't forget to check out the promotional material that you can find in the docs tab. The the column is on the right hand side above the chat box. So there's a great ebook in there on introducing AI and two handy documents looking into planning of the supply chain future. You can also find the SAP Discovery Center in there. So don't forget to check out those. We will now move to questions. I can see loads been have been coming in as we've been talking. So we've got the first one here from Kartik that says, what data governance, cleansing, and preprocessing steps are critical before enabling embedded AI, and does SAP provide automation or tooling for this? So I think I mean, one could talk a couple of hours just about that, but just to trying to be brief. There is, of course, as as we said before, a little bit of a difference between the traditional AI stuff and the generative AI stuff. Right? So for the traditional AI, when you're talking about the transactional data and the master data, So in IDP, for for example, we have some machine learning for the master data to fix issues in the master data. We have machine learning for the outlier corrections for transactional data. And so those are tools you can already directly use embedded in ADP for the more like, for the demand forecasting, for the optimizer, for for the existing embedded stuff. And that's something that, of course, you should and can be take care of. If you're thinking about broadly more broadly speaking about including external data, that's not only, you know, planning specific data at SAP. There's also, you know, on on BDC now, there's there's Databricks that has plenty of tools. We have the I foundation on b d BTP, which has a lot of machine learning tools that can help with that stuff. But that's all and that's only the traditional AI piece. On the generic AI side, actually, for most of the embedded things, you don't need to do any preparation because that's instead using kind of the internal data that the system already has to to create some some some natural language out output. And but as I mentioned, and and as we discussed then with with Jewel and in the future with agent, it will be about, you know, writing down the processes and asking the questions the the right questions in a sense. And but that's and so that there is no cleaning, I would say, for that. Right? It's more a question of getting used to the new way of interacting with the system. So those are the two areas I would consider. Maybe I don't know if you have anything that I forgot. No. I think those are the two big ones. I think that it's you know, the the you know, from supply chain planning, it's fairly compartmentalized. Right? So all the data for our from our perspective, all the data is in our platform. And as Laura noted, if if we are making sure that you don't have bad model data from a supply chain perspective, that means that all of our solvers do better, whether it's, you know, AI optimization or or any of those. And so I think that's a that is a big piece, and and we do have tools around that. Great. So we've got another data question next. So Amy has asked, any suggestion on how to collect scope free data from SCM data? Well, I will have to admit that I have to Google what scope free data is. But it seems to be like emissions data. Right? And I don't don't believe we have any automated way to do this. But I believe there was also someone answering partially in the chat. We, messaged can do, it. I can comment on that. Is that so we do have within SAP integrated business planning, we do have carbon roll up information that. can be that they can may maintain. And so that that drives some of that. Right? So we do have that ability to do that. That's part of scope three is my understanding. And so so there is there is pieces of that. And then some of it is simply that, you know, you take whatever plans you have with volumes and, you know, some of the activities, and you may have to enhance that as well and potentially through other reports. But but but the the carbon roll up does exist, and that the other would be extracting just volumes and of materials that you're going to produce, and then you may have to enhance that as well. Yeah. We've got another question here from Thomas, and it says, do the agents include the opportunity to include learned attributes make self correcting adjustments, or is that still a human based input? Well well, that's the goal. The first version will not be self correcting. Right? I mean, we are realistically speaking, that's still going to take some time. Right? The first agents that are going to come out there are going to be agents that that have already predefined instructions, and then you have to manually change the instructions if you want them to do something differently because, you know, some parameters learn things. So there still needs to be a human component. And then, of course and, you know, the the the self learning piece can only happen with time after you've also collected all the data, you know, over and over again, how the agent behaves, what actions is it taking, which of these actions were good actions in the opinion of the humans, the of the planners. So you have to collect all these data in terms of what did the agent do and how good did the planner think that the result was. And after only after you start collecting the data for a while, we will be able to actually then have agents that self correct. Right? So it's it's going to be a process. I I would note I would note also that, within supply chain planning, fundamentally, I think that there's a challenge of deciding what a good decision is. Right. And so this is true with planners today is that if you look at most companies and if you were to ask them, do you do you measure your your planner? You know? Yes. They measure the planners on maybe forecast accuracy, so the end end point, but that's not that doesn't get into specific decisions they made that drove accuracy as well as supply planning is the same that might be scheduled adherence. But, again, you're talking you're not talking at discrete decision level. And so I think there's a lot more data, as Laura talked about, for us to really start trying to figure out how do you how do you get that. The starting point might just be, as Laura noted, having having planners note whether that was a good decision. Right? So you'd have to have planners go through not just the decisions that the agent needs help on, but also evaluating here are the 20 decisions that the agent made. Do you think those are the right decisions? And so that you have to find that, but I I think it's always a challenge in supply chain as to was something the right decision or not. It is fundamentally a challenge. And there will always be that time that you have to put into a new AI agent where you have to nurture it and teach it. And so that we were saying earlier, you know, teach it how you how you speak and how you think and what. your company needs. So we got one more question from the chat chat here, and that's from Michael. It says, is it possible to use embedded AI agents in SAP ECC six version. I don't know. I I I the answer to that, I think, is not from an SAP perspective. Right. And so at at the I don't wanna say no because there's a whole a whole range of APIs that are available in ECC six six point o. But SAP is not building agents and embedded agents to use with ECC six point o. And we just had one question pop up, and it says, I still have concerns on safety of the data of information that's being exchanged with AI or agents. Any thoughts on that? And that's the top of the main questions, I think. And nowadays, data privacy is is for sure. It's a challenge, but I think so at the end of the day, it will depend again on the scope of the agent. Right? And and I think, again, if you think about the agent as a new hire, the new. hire now have the same permission permissions and authorizations of an existing employee, which has a specific scope of work, and then it's only allowed to read and write exactly the same kind of things that the existing employee can. Or it's going to get its own personalized permissions and authorizations around data because their scope is a little bit different from what existing employees are doing. And so there's still going to be very much a topic of permissions and authorizations for agents that's not gonna wear. Like, the same kind of user authorization topic that you have for employees, you have to imagine doing the same for agents. And it will depend on exactly what the agent is supposed to do. So for sure, there needs to be also a bit of supervision from also from customer perspective to make sure that you only give the agent the rights to the important things and and and and nothing else. From SAP perspective, for sure, we are looking into, for example, agents are supposed to produce steps Planner would do, that the agent kind of inherits the same kind of authorizations and permissions that the planner would have. So what are some the guard like, the guardrails that we're putting in place to make sure that AI, you know, operates within and it doesn't go off and make its own decisions and hallucinate? And how how, like, how are we controlling the AI here? Well, It's mean, a very a very broad, but it's I mean, at the end of the day, as I said, right, so agents are going to be a set of instructions instructions with a set of tools. And you predefine which tools they are allowed to access, and within these tools, there are guard lays guardrails, like, you know, they are APIs to extract data. Oh, well, they can only extract this and that type of data and not that other type of data kind of thing. Right? So there is multiple levels to it. At the end of the day, it's a combination of the guardrails on the API side, so the tool side, which access they have to which tools and how much these tools can be allowed to do things. And it's going to be a combination of that and how well you define the instructions. If you have very well defined tools, but you're super vague in the instructions and you let the agent just go wild, that's also going to be a problem. So you will need to keep an eye on both side of things. Great. So, yeah, a it's it's an error, user error. I think the other thing that is is pretty common is people's concern about large language models and data being sent to them and used as training data. And I think that, you know, when you think of agents, that there's multiple pieces to agents. Right? And and one piece of it might leverage an LLM. And if somebody is triggering an agent and they're they're talking, we need the LLM to understand what that what the the natural language is. But we don't necessarily need the the the LLM when we're operating on data that is within our system, so within, like, SAP integrated business planning. Right? So if you if you are triggering the agent, then there's a whole bunch of rules and everything. The LLM is the only piece that might might have any external date any data sent to it and and brought back. And if that is simply just I just need to understand the terminology that's being or what is being asked, and then we parse that in and then everything else is through APIs, as Laura said. So I I think that that's one of the things that is, I think it's a challenge for everybody these days to think about, you know, when when you say AI, everyone has their own conceived notion of what AI is and and what their concerns are, and not all AI has that that concern. And so I think that's a challenge for us as a software provider to let them know what we're doing and and where there is data that could be sent out, what data is sent out. And SAP has a whole framework for this or Ethical AI framework on, you know, what is allowed to be sent, what can't be sent, you know, how do we how do we deal with a lot of these issues. So I think we've got time for one more question. They are coming in quick and fast. I'm sorry if we haven't got around to your question. And it's, what are the various AI use cases which are embedded in SAP extended warehouse management platform? And are these AI agents being rolled out in Asia and India? So one of the links that we sent should have been the SAP Discovery Center. So there is SAP publishes for anybody. You don't need to log in. You don't need to be an SAP customer. Anybody can go look at it. And if you search on Discovery Center SAP, you can get a list of all of the AI use cases that SAP delivers. You can also sort that and filter it. You can filter it by the product. So in the case for extended warehouse management, you can find SAP extended warehouse management as product, and it will show you all of the AI use cases. It will also show you whether those use cases are considered embedded or, you know, basic AI or premium AI. All of that information is available there, and I think it's probably the best resource for that. I would also note that we generally don't we don't roll out AI AI agents or AI features by region at SAP. So in general, if it says it's available, it's available everywhere. Brilliant. And one final piece one final piece of advice, really. If if someone is just about to start their journey using AI, what where where should they look first? Where should they start? I would well, I so, again, there's AI is a big, huge, broad subject, and we've talked about a couple different, you know, use cases and the amount of effort needed. But, honestly, if I were to start and say, you know, if I have the mandate from management that says you have to use AI in your jobs, I would start at SAP, I would start by connecting Juul up to your integrated planning system or whatever system you connected to, and that allows you to start using natural language interaction. So you're using AI. You know, some of the ones that Laura talked about, some of the traditional AI, forecasting, that's more of a project. Right? You have to decide what data, and you might need some other things. And so that's a little bit bigger. But but and then we've also got some use cases around some of the explainability that says you're are if you're already running SAP integrated business planning, you're running a forecast. If I want to if I want to start using the the the analysis of my forecast, that's the button you press. So there's you really have to look at all different use cases, and and some of them are very easy to consume. Some of them are more of a project. So I think that's the way I get it. And I I would do a little bit of self advertisement here. The the ebook that's then linked in the docs tab, that's, you know, that's a practical practical guide on how to start using the the existing AI use cases in IDP. So both dual and the existing, like, demand forecasting, for example, how to set it up, where to look, and so on and so forth. But, of course, you also can find much of this information on the SAP help. So it's it's, you know, it's out there, the information, and either piece it together or have a look at the at the book. Well, thank you so much, John Laura. That unfortunately brings us to the end of our discussion today. We've seen today that future proofing your supply chain requires a shift towards intelligent embedded solutions that actually empower your people. Before we go, head on over to the supply chain digital website and check out the webinar section for more industry insights. And once again, make sure you grab hold of all of that good stuff that Laura has just mentioned in the document tab. It's on your right hand side next to the chat box. Thank you again, Laura and Todd, for joining me today, and thank all of you for attending. Laura and Todd are also happy for you to follow them on LinkedIn. I did check with them and their five of it. If you've been inspired by this webinar, to do so. This webinar will be online shortly, so come back and watch it again and again. Remember to share it. But for now, goodbye, and see you all soon. Thank you. Thank you so much.