Name: AI in Supply Chain
Uploaded: 2025-11-07T18:16:08.198Z
Duration: 1 h 2 min
Description: AI in Supply Chain

Transcript for "AI in Supply Chain": Hello, everybody. Welcome, everyone, to the Real AI in Supply Chain Academy show. We are having a lot of people join in right now, but, in the interest of time, let me get started. My name is Nari Viswanathan. I'm senior director of product marketing at Coupa, and I've been joined by or I'll be joined by a whole host of really great speakers to talk about various topics within the realm of supply chain. We're very thrilled to have you join, us in this particular, area of how AI works in the realm of supply chain. Because supply chain is becoming lot more strategic in organizations and bringing in competitive advantages amidst constant disruption and pressures. So it's very critical for us to implement AI in this area. So before we get into the topics, let me just briefly talk about some of the quick logistics. If you look at this particular, platform, there are a chat area where you're welcome to provide your inputs and observations and say hi. In in fact, you can let us know which location you are joining us from. But if you'd like to ask some questions, there is a q and a box. So, please, feel free to send in questions in the q and a box, and we will be able to answer them pretty quickly. So here we have the speakers for today. As I mentioned, my name is Nari. I'd like, to, introduce our panel of experts. My role here in this particular webinar will be to moderate the session. And then we'll have, Ayushman, who is the digital supply chain leader from Deloitte. He'll be providing us with some key industry perspectives. From Coupa, we have Gaurav Goyal, who is the director of product management for this particular solution area. We also have Lucas. So Lucas DeBrito, who is going to show you the software. Is the principal solution adviser. And I'm also excited to see the latest and greatest demonstration. So in terms of the agenda for today, as I mentioned, we'll start off with an industry perspective from Aishman. Then we will talk about kind of the architecture and how our solution works in tandem with optimization for the Coupa, Navi supply chain agents. And then finally, we'll focus a lot of time most of our time will be on showing you the software. And then as I mentioned, please, send your questions in the q and a, area, then we'll try to, answer all of those questions during the q and a, time frame. But before we get started, we just wanted to do a quick poll, and we really wanted to see, and find out how would you classify your own level of AI understanding. Would you categorize yourself as a beginner? Would you categorize yourself as an intermediate or advanced? So let's see how, you know, what how the audience feels about that. So the poll is ongoing. And You see that Aishman is ready here to get started. Welcome, Aishman, on stage. Alright. So it looks like most of the audience is, in the beginning stages of their journey towards AI, which implies that I think this session will be very helpful to kind of get the perspective from Deloitte, because I think they have a vast array of use cases and vast array of customers that they work with. So I think that that will be very beneficial for this audience. So with this, I wanted to hand it over to you, Ayush. Great. Thanks, Nari. Good morning, good afternoon, everyone. Quick introvert myself. I'm Ayushman, working with Deloitte for past eleven plus years. I work on the intersection of supply chain and digital transformations. We've helped multiple clients kind of, first kind of define their vision specifically around digital transformations, specifically around Gen AI and its impact to supply chain and how we can use those things. So, happy to talk about it here and and give some real world examples of how to apply GenAI and agentic AI into your supply chain processes and supply chain solutions. But before we start on that, I wanna talk about how has AI evolved over what we see. Right? First of all, AI enabled models have been around here for a while. Right? But, the big bang movement for AI came in 2022 when Chad GPT introduced to us the concept of LLMs. Right? Large language models. Right? That was the original concept where they could have some, you know, they would have, they could chat and interact with the chatbot. They were able to get the context. They were able to predict the next words. They could give give us some ideas, and they could generate content. Right? That was raw generation, whether it was text, whether it was images, whether it was audio. Right? That was back in 2022. It gradually got better. We introduced RAG, which helped it to give it a bit more context, bit more memory. Right? That helps to reduce a bit more hallucination that we were seeing in those that were there in very early stages. Right? It was now helping us kind of summarize documents. Right? It could get that context. You could feed the LLM multiple documents, and then it would kind of summarize it for you. It would highlight a few things for you. Where Bolt has shifted since then, and this is where we are kind of currently living in 2425, is we have now moved on from LLM and RAG to build certain simple agents. Right? Now AI can help you plan, automate. Right? It can help you validate and update the documentation. Right? Whereas earlier, it was just kind of summarizing for you. It was giving you its output. That's it. And then it was that that was just the transaction. Now AI can kind of give you predictions. It can give you inputs. And then human in the loop kind of corrects and, then can give them feedback. So it's kind of an iterative loop now. Where we are seeing now as we look forward is the world and the world of supply chain shift is we are seeing more and more clients are looking towards multi agent systems. Right? Where we are not just working on a single use case, but can we work on the whole workflow together. Right? So multiple agents kind of talking, running together, interacting together, making those kind of inferences from the past data, from your input, and then also talking to the user using a chatbot, right, or using your Teams or a Slack or a communicator. Right? So and at the same time, probably talking to an external provider or an external stakeholder as well. So that's where we are thinking and seeing the marketplace shifting and evolving. And then, again, looking ahead again, I'm not a I don't know a crystal ball in my hand. But as I look ahead on where does this world look like, is a lot of clients kind of are looking to build their own agents, like the DIY agent that we are calling it, which is basically agents configured to my use, to Nari's use. Right? Again, I do this piece of work. Can I build my agent to run my own kind of supply chain transaction? Can I build my agent to run my marketing campaign? Right? So that's where the world is shifting. You're getting a bit more personalization. At least we foresee people will kind of build their own agents, customize their own agents. And then, you know, this is the journey of agent DKI. Alright? Let's move to the next slide. Right? So, now what are we seeing in the marketplace? Right? So Deloitte did a CPU survey and we want to understand, hey. What are our enterprise priorities, coming in from the procurement organization which are critical to our supply chain. First and foremost, no no surprises there that the first top two priorities are always from a supply chain perspective is improving margins, cost reduction, and then operational efficiency. You can never take that out from a supply chain professional. But the third most biggest factor that the supply chain professionals and the procurement professionals are looking at is digital transformation, is AI. Right? All the industry leaders are trying to look at, are there AI solutions that will work for them? Are they fit for purpose? Are they purpose built? They don't just work for them today, but they work for them future for their growth. All of them are looking to invest in digital transformation and AI. All the procurement leaders, if you see by data here what we gathered, this survey was conducted, I wanna say, q two or q three of this year, so pretty recent, that we had procurement technology as one of their biggest priority. Right? Any additional budget that the procurement leaders had, they recognized that GenAI was their top priority to invest that budget to kind of learn more, understand about AI. Right? And they are likely to more use those next gen technologies where earlier people were not so hesitant to kind of, you know, use those. They were they were waiting for tried and tested technologies. But with the advent of AI, we are seeing a lot more supply chain and procurement transformation procurement teams looking to do some pilots and POCs to understand how can I make AI work for myself? Right? So where does that take us take us? Right? Let's jump to the next slide. So what are the various use cases that we are seeing overall end to end across supply chain, that we are looking at? Nani, let's jump to the next slide. So what we have done is we have split those, AI use cases into three buckets. Right? So one is standard system provider agents. Right? So you already have technology solutions like Coupa Software, other players in the market. They will also already are coming up with their own agent, and we'll see a demo of it, in in a short while. Right? So so those solutions, if you have employed that system correctly, if you've kind of done the due diligence, done the basics, you will have those agents based on the technology providers that you have. The second set of agents that we think can go slightly broader than that is a use case agent that's working on a given use case, and it can go beyond a particular pillar of supply chain. Right? And then the third one is more around decision intelligence agents, which go end to end, work across the supply chain. They're able to connect the dots. So it's it's something similar to those multi agent to DIY agents group that we talked about is how they can, you know, interact with different agents in different areas, within different technologies, go across systems, and then kind of build your agents. Right? Let me kind of touch a couple examples. I don't wanna play in this slide. But from a system provider agent perspective, you could have, an automated task execution agent, right, that you build some logic during your plan or during your sourcing or during your procurement phase and say, you know, this is my task. Just make sure you check the relevant data, do your validations. If I'm submitting a PRPO or a contract or an invoice, make sure they do their validation. That's a task revision plan. It will work within that system. Right? The next phase could be, if I wanna do some spend, you know, spend checks and balances, if I wanna build some reporting, That particular agent from a use case perspective needs to go across systems. It needs to go into my planning system to check, okay, what was my budget? What was my plan? It needs to go into my sourcing and procurement system to understand where did I deliver on those, you know, plan? Did I go above? Did I go less? You know, what is my run rate? Do I have enough? Do I have, you know, for my month end close processes. Right? So if I'm gonna build that sort of an agent that will help me track my spend, track my budget, it goes across, you know, supply chain pillars, it goes across systems. And then the last bucket is agents that kind of leapfrog across the whole system. They kind of go across the end to end supply chain. Think of it something like a control tower, like a single pane of glass view, which will bring in data from everywhere and then we'll run that agents on top of it, give you pinpointed feedback and recommendations, give you alerts, give you real time adjustment that, hey. This particular KPI or SLA is turning red. You know, they could give you a recommended action as well. So that's kind of the decision intelligence agent which can make those decisions on your behalf. Obviously, human in the loop still exists. At least that's what we are seeing in the market that even though you could get the confidence level up to a very high degree, all the clients in the marketplace still prefer human in the loop till the time they can automate all those agents. Couple of agents highlighted here in pink, as you can see, is around demand supply balancing and the network design. We'll touch upon it in our demo later today, by the Coupa Software experts on the call. Alright. Let's jump to the next slide. Right? So now what we've seen again, it's all hunky dory. I mean, now we are talking. Everyone can go ahead and apply AI. Is that the the real, you know, is that the the real world implication of it? That's no. We are seeing some challenges that we've seen from our clients, within the AI world as well. Right? Again, I'll not speak to all of it. All of them are self explanatory. But couple of them that I wanna highlight, which are very clear, with that that we are hearing more and more from our clients. Like, obviously, data quality, talent, we've been hearing that in the marketplace as well. But then having a clear understanding of what's my business case, what's my ROI, what is the use case that I'm trying to solve is critical when you're going in for AI adoption. It's critical when you're going in for a pilot or a proof of concept that you wanna do for your AI adoption. And then another thing that you've seen client asking more and more like a like I said, still till now, I think there's still the final review or the final step will always be like a human in the loop. Right? But now clients are asking, can I get more trust? Can I make it automated? And there's also other side of the coin where clients are saying, yeah. I don't trust my data, so I don't trust the decision or the inference or or the recommendation by the AI model. So there's an aspect of training your model correctly and the trust level that you have with the model that we are seeing clients who implemented AI kind of, you know, getting to that next level of how I can make it better, how can I make it fully automated, you know, without human in the loop? But then that factor of model, data, and trust is very critical. But, again, couple of the factors that I wanted to highlight, like chain management, ethics, all of those are important, for us to, you know, tackle this AI, you know, as a solution that it works for us and it works for our organizations, for our, companies. Now how are leaders adopting it? Right? In despite of the challenges, how are we seeing other leaders, kind of get through those challenges? Let's do the let's go to the next slide. Is there there's two buckets of it of what we are seeing in the marketplace, again, coming from our CTO survey. There's an aspect of industry leaders who went in AI, kind of kind of, you know, double dip on it. They've kind of done some proof of concept. They have done some pilots. They show a much better understanding of the AI. So what 53% of our survey respondents show that they had clarity going in. They've they understood their business case. They understood their ROI, what they were solving for. Right? And then what is the common strategy to do that? It's kind of dual fold. One is, I don't have talent right now. I don't have technology right now. Let me buy the best of breed. Like, let me get Coupa Software to help me out here. And then there's also an aspect of clients we're saying they have, you know, their reservations around their data, they want a private LLM, they are spending more time. It's a slightly longer term view of developing a GenAI and an LLM board in house. Again, those two are the kind of the two major strategies. Again, depending on the priority, the, you know, the ROI that you can expect, the data sanctity issues that certain clients have, they are using those methodologies to get well versed with AI and let AI give them the results that they want want. But there's also an aspect of, you know, in contrast, what 3% respondent, which we are kind of categorizing as followers, they're still kind of in planning and exploration phase. They're kind of laggards off that layer where they're still moving slowly. They're still in uncertainty. Right? So I think our recommendation to all those clients are at least, you know, the the you cannot just be sitting on the side fence. This is time for actions. Right? So go ahead and do a pilot. You know, work with work with the providers in the marketplace, work with your technology providers, work with your, you know, solution consultants to understand this EIs. It's gonna be here. You have to work with it. Don't just sit on the fence. Make a move and, again, it could be a small move, but it could be a pilot or a proof of concept. Right? And then how do we get on with that strategy? Like, a lot of clients come to us and ask, hey. I don't know. How do we how do I make a move? How do I get started with my AI journey? So how do we recommend that? Let's move to the next slide. Is, it's a very simple framework that I recommend to my clients on how do you kick start your journey. Right? The first thing that you would always do is know your possibilities. Right? Like, what is the out of possible? What can AI solve for you? Like, a lot of clients just think AI is chat gbt. It's a chatbot. I just chat to it. It just give me answers. Right? There's a lot more that you can do with the AI. You know, so it's understand that are possible. Do you wanna do that all all those things on day one? No. Right? But under at least understand what your vision is, what the are possible is, which will help you define your organization, your business unit's vision. Right? Once you've defined that vision, understand your use cases. What are my gaps? What are my problem in this? Right? I wanna solve those specifically. Right? Prioritize those use cases. Right? And then you can one step kind of, you know, activate your vision. Right? Identify the technology partner. Do you wanna do this in house versus do you wanna do it with an external provider? Prepare your data. And then, again, go in a pilot format is what we've always recommended. Do a proof proof of concept. I trade over it. You know, I trade over improvements. The first scenario that you deploy will not be great, but as you I similar to how you do it with a chat GPT. Right? The first comment that you ask is not great. But as you give it more context, as you iterate with it, as you talk with it, you gradually improve and enhance. And then the foundation on which all this lies is the tech stack that you build together, make sure your architecture is in sync, the data readiness. Needless to say, data is all all all the more important now. And then the chain management aspect of it. Make sure that your organizations, your teams are ready to change as you change along with AI. So that's all from my side. Nadi, back to you. Yeah. Thank you. Thank you, Aish. We actually received a couple of questions which pretty much was in sim along similar lines. More about, like, the duration of some of these pilots. You talked about the fact that, hey. It's important to get started Yep. To a pilot. How long do some of these pilots take? So depending on the scope, we've seen certain pilots that can go very quickly, like, in a quick six to eight week, ten week period. Versus if you're doing an enhanced pilot, you kind of phase it out, MBP one, MBP two, they could go on to, like, eighteen, twenty weeks. So then depending on the use case, depending on whether you're doing it in a small geography, you know, small country, small BU versus a global organization, you could take that phased approach. So, again, I don't know if it was a proper consulting answer or it depends. But depending on the scope and the use case, depending on the size of your organization, that's what we're seeing. But yeah. But those have been successful. Like, we've been organizations that did a small pilot of six to eight weeks, understood their weaknesses, like, say, the data was not in place, if the tech stack was not in place, and then use that to do the next round, which was another six to eight weeks, then iterate and improve over it. That's what we've seen in the marketplace. Yeah. Make makes sense. In fact, we received one more question about the implementation teams, receiving training in AI concepts so that they can guide customers on how to leverage AI features during implementation. Yes. Absolutely. Yes. In fact, when we show the demo, you'll see that it's very intuitive in terms of how, you know, anybody can just go ahead and start using it. But from an implementation team, of course, they they have to know more details of how the underlying, capability works, in terms of where it's going to get the data and things like that. So, yeah, so they will be trained both for our supply chain design and planning as well as for our procurement solutions that they'll be training. And then the training will be kind of twofold. Right? One is, obviously, with the technology that you're bringing this new techno how this works, you know, what is the core config behind it. But there's also an aspect of what does this mean for my process, What does this mean for my end to end supply chain? So so be ready to kind of invest and implement chain management and training there as well. Yep. Makes sense. One more question. Ayish, how is the industry measuring the ROI, cost of goods sale, savings? You know, what kind of metric is used? Typically, what we've seen is, again so if I implement a particular AI bot and if it helps me, say let me get an example of legal reviews. Right? Like, I'm doing a contract and I'm spending so much time with my legal reviews. And if I save time using an AI bot that can summarize my contract, it can give me pointers that, hey. This is a mismatch or this is a risk area that that the legal team needs to focus on. So instead of legal team kind of scrolling through tons and pages of the whole, you know, contract, they can just pinpoint into the clause, the paragraph, the area they need to focus on. Right? Now let's say they took two weeks to review that earlier, and now they can do it in two days. Right? So we are seeing the investment that we made to build that one particular use case for my legal team, And now we are seeing the benefit is we save time for those legal teams. Right? That's gonna get from a sourcing and procurement side where we've seen, legal team saving time, and then you can quickly calculate, you know, the benefit that you got versus the investment that you did. Agreed. Yeah. That that makes a lot of sense. Okay. Cool. So I think now we are in the next, section, where I'll request Gigi to talk about, kind of the underlying capabilities or the technology behind the Coupa Software Software supply chain Navi agents. Thank you, Nari. Thank you. Hello, everyone. I'm very excited to say that Coupa Software recently released, Navi, an agent tech AI system, specifically purpose built for supply chain needs. So I'm gonna be just walking you through what is it exactly means, how we will make it, like, for supply chain use cases, and specifically also covering on how can we make it hallucination free. Like, let's go to the next slide. Perfect. So let's go to, like, you know, some of those things which we keep hearing in the market. Right? A like, you know, supply chain is very complex. We all know about it. Right? It's very, very driven by data, and data is all about analytics, which we do before we do any sort of planning or modeling work. Right? So when you put Gen AI on top of it, it actually fails significantly because Gen AI, the nature of it is generated. Like, it generates attacks. It generates the images based on what he has learned. It doesn't understand the nomadic or the numbers behind it. Right? The mathematical reasoning is not being built into it. So what we have done specifically is as a differentiator is we have put a reasoning at or mathematical reasoning, if I may say that, at a scale in in our, Nava iden tech system. Essentially, it knows, specifically this is a supply chain use case, and it also knows that a two plus two is always four. It will never be 4.1 or five. So that is what specifically purpose built supply chain, Nava agentic system is. The other thing which we have made pretty clear about it is we have trained it on a structured data. So, essentially, when you look into any sort of again, going back to traditional AI assistance, you will see that they are being done on a lot of unstructured data. Right? A lot of the text finding does and data mining goes behind it. But we have to make sure, again, that it is being designed on a specific schema, specific fields, and values so that it is very deterministic in nature. Again, going back to that, if you ask your question two months later, the same question, if you repeat it now, you will always see the same response coming through it, provided underlying data has not changed. So repeatability is also one of the thing which we have tried to build significantly in Navi's agenting system. The third key differentiator you will see here is it's an AI led vision which we follow in Coupa Software. So, essentially, now Navi is sitting in the center of all the product portfolio. When we look into the Navi, we look into that how can Navi interact with different products, and how can that that data harmonization, data integration can work together when we we use a user's Navi. So, essentially, it's not a cherry on the top sort of structure or holding, another layer on the top. It's actually very well embedded into the product portfolio. The, next differentiator, I will say, is the seamless integration. Essentially, when you talk about, Navi, you can have an end to end workflow being consumed by Navi and part of our mission. So, essentially, when Piyush does Dante, hey. Workflow automation is one part of it. Bring your own agent is another part where customers are looking or clients are looking in the future. This is where our vision also is. We are looking for a word where users can bring their own agent into our product portfolio and then start interacting with the other products which align into the product build. And last but not the least, as Coupa Software is known for, we have an enterprise grade security and privacy for other products. Same goes with Navi. We never share any data with any NLM provider or any AI providers per per se. All the data is very much standardized, normalized for the specific use cases and very much reside within the Coupa Software. Even we, as in Coupa Software, don't get to see the data in, the real fashion. We only get to see into the patterns and, most of them, it's fairly a, pretty much encrypted dataset. So this is how we think that supply chain definitely deserves a specific, agenda system, and we have purpose built that for supply chain. Now you can go to the next slide. Perfect. So how do with all that thing, how does the customer benefit from it or a user benefit from it? First of all, Navi is built in a way that it can work across the paradigm of the supply chain. You start with design, but you can go into planning. You can go into the execution. The more use cases you add into it, the more workflow automation will it will be built on to. So, essentially, a place where a new and time you can go into as a user, and you can start interacting. And it can let it walk through a lot of hoops within your portfolio and get you the best answer coming out of it. So with that, the first benefit we have seen in our, you know, early user testing is ramping up for the users. The users who are very much new to our product portfolio were able to ask those questions to Navi and quickly get the answers without looking into the help documentation or asking someone else for some clarification. They can quickly ask Navi, and Navi was able to specifically pinpoint where you can do what you can do. The other thing, Spectrum of the users were very advanced users. Like, they were on top of everything. They wanted to use most of the features as possible. For them also, Navi came pretty handy when they asked those advanced questions and that was specifically, oh, yeah. This physical feature was recently released, and this is how you can use it in your supply chain, use cases specifically. So huge productivity or or reduction of ramp up time when it comes to the users. The second benefit which we have seen is intelligent trade offs. Essentially, when you ask Navi to tell me the summary of the scenario, it gives you a pretty detailed summary and then also give you a good insight about how does it compare with your base line. So you can see that which KPI is improving, which is not getting to the zone where you would like to be comfortable with. So all the trade off divisions, Navi can take out for you. All you need is to ask. Just give it a prompt, and Navi will give you all the compatible analysis, on basis on the KPIs for you. The third thing, personify Navi. Essentially, when we looked into Navi as one of the product, we said that Navi will have a different personality depending on who is asking that question. A person shipping in the Swiss, we might not want to go into the features and functionalities of Navi, but would like to know what supply chain p and l impact, on the overall group. Right? So in that case, you can always ask Navi, hey. You imagine you are a financial geek or you work in a financial, portfolio. Can you help understand what the scenario is for me? So, essentially, Navi will mine through all the profit margins, revenues, EBITDA, all of that stuff and give you the specific concept related to the finance. The last but not the least, whenever any incident happens, we are building a portfolio where now we will be able to specifically tell you what really has changed from last time you ran it or last time you used the scenario versus now. So you can specifically go, hey. This is the factor which is driving the change in my supply chain, and probably this is the area I should go and tackle it first. So these are the some of the great benefits which we have seen our users are having while they have started using Navi, like, from an earlier doctor program to once it went well GA. Now you can go to the next slide. Perfect. So from our vision standpoint, right, our vision when we started looking into the Navi, hey. We should be a productivity and efficiency multiplier. Essentially, what you were doing and taking for, like, as Ayush mentioned, like, it was taking couple of days to turn into hours, right, from an even standpoint. But the same is for supply chain. We want to do let our users do more with the time they have. Right? And not only that, we want them to do, not all the things which they were not able to accomplish earlier. We heard so many stories from our customers where they always wanted to do they knew that there is some sort of, you know, golden nugget lying over there if I dig deep into this data, but they never had time to get there. Right? Call it geopolitical climate or economic issues, which we are dealing with, but they never had time. But with Navi, they can ask that and let Navi do the work and find that information for you and get everything what you need for. Right? So what we think of Navi, specifically, from a vision standpoint, will stand on four pillars. The first one on the left side is essentially knowledge generation, which is essentially the community gen AI piece. Hey. How can I do it? Where can I find it? So on and so forth. But we are actually building on top of it too while we have that. We are also building a guided user workflows. Right? When you ask Navi a question, Navi will take you there directly and help you out. And this is a specific table. This is a specific column you should look into when you are looking to change safety stock for that example. Right? That is the one piece we started dealing with, and Lucas will show you where we are on this journey. The second one is about user workflow augmentation. Like, help me out understanding this gigabyte of the data. Let me get mine through this and help me out so that I can know I can describe go to the description state from the baseline and come up with certain sort of scenarios to work with. So Navi helps your users to summarize their some, their scenarios, compare between different scenarios, different parameters, and explain the results what you are seeing in it. For example, if you ask Navi, hey. My scenario is infeasible. I'm not able to run it out. What are the causes of that? And it can pretty well give you the detailed list that this is the things which are not, working or the constraints are being hit, and that's why your scenario is infeasible. While we have that as a foundation, we are going towards the further right side of the equation. Right? This is essentially workflow automation. All these agents were which are super specialized with specific tasks will be orchestrated by a central layer, which will help in orchestrating each of them. Imagine a word you ask, hey. My demand has increased by 10%. Tell me what should I do. So for that, now you will understand, hey. If you wanna, hey, create a scenario item to increase the demand by 10% on a specific location, create a scenario, run that scenario, analyze the output, compare it with baseline, and tell you the, results out of it. All of that can be done end to end automatically with with this agentic system which we are building. And the last but not the least, AI control plane. We talk a lot about what users will touch and feel. This is all about what in the back end we will accomplish with this agentic system. Essentially, when you ask Navi a question, you will get, I would say, estimate time to accomplish that task. For example, if you ask Navi a question that I want to run this scenario, and Navi will come back and say, hey. Running this scenario will take, like, four hours, and you probably nudge Navi. Hey. I don't have four hours. I have to finish that in one hour. And Navi will go back and say, okay. What all could be done to get it run-in one hour? And it would give you the clear dog decisions you have to make. For instance, it might say, hey. Relax this constraint, and you will be able to reduce your time by 40%. Change this lane to this lane because the policy allows you to do that. You can further reduce it by 20%. So then user is in loop, and then we'll say, okay. I have two one and two. Navi then comes back with that specific, new revised estimate and then runs that. So, essentially, all that intelligence with course in the back end, we are also plugging it as a part of Navi architecture. And, again, this will be available across all the technologies which Coupa Software has, network optimization, transportation optimization, inventory optimization, and simulation. We can go to the next slide, Niles. Perfect. I think all of this is pretty fantastic, but the thing which always keep everyone awake at night, how much is this? Once I started using Navi, how much can I trust it? Where is where are the things what if the things doesn't work the way I want it to be? Like, data is not being created or data has not been analyzed the way I was expected it to be by like, the human will do. Right? So hallucination is the word which is going in and around the genetic AI. And I told you, right, Navi is is an agentic system. It has generative AI as one of the component, but that's just one of the needed in the mix. We have other technologies also coming into the play. So what we have done is we've come up with a composite AI techniques. And imagine you have these four entities working in in series here. Right? For a simple example, you have human who's intent the prompt. There's an agent which is understanding the prompt. And a lot of them which is running behind the scenes and then it's mining to the data. And, of course, the, opposite flow happens after that. So these are there are the, I think, four key techniques which we are using to make sure it's hallucination free. First of all, depending on the prompt which user is asking, we are dynamically selecting the agent to do best to execute that task. So, for example, if the task when you prompt leanings towards creating a scenario, it will never go to implicit diagnosis at the 1st Floor. It will go to a scenario creation agent, and then that has to accomplish that task once the output of that spits out, then some other agent will pick it. So dynamic selection of agent algorithm data that is happening, to make sure that we are getting, to the right level of accuracy and the quality. The second thing which we talked about is everything is the word is neurosymbolic, but, essentially, it means everything has a logic built into it. For instance, if policy table is linked with master data and then policy table is also linked to the like, production policy linked with inventory policy, all of that linkages has been set by the architecture, and they are the fundamentals to the supply chain. Right? Only supply chain practitioner understands that, and we have we let AI learn through that, Nava learn through that so that it understands and wants the supply chain and talk as a language of supply chain too. The third is about iterative complexity management as you're seeing here. But, essentially, the idea is if you ask a complex prompt too much for an RB to comprehend, it will actually, in the back end, break it down into couple of sub problems to deal with so that it doesn't jumbles up and come up with some random answer. Right? So, for example, if you ask, Navi, hey. My demand is again, going back to the same example. Demand is up by 10%. What should I do? It actually has couple of tasks involved in it. Right? First, create a scenario item, check if the scenario item is available or not, then run the scenario, then create the output. All of that has been divided into sub problems, and then series of those tasks will be accomplished before you see the output. And finally, the last but not the least, we are coming up with a way there all the output which we got from the different different algorithms or the agents are being vetted by in a central body. Essentially, there is a score to the solution quantity when user sees, or where the agentic system sees that solution quantity reaches the threshold, then only you will see able to see the particular output. Otherwise, it will generally come back and say that, hey. I'm not very confident of solving this out. How about, you know, give me more information before I can respond to you for clarification standpoint? So by all these measures, we are trying to contain the hallucination because we know our supply chains can't run on hallucinations. Supply chain can't be run, without understanding the data underneath. And I go to the next slide. Okay. Perfect. Now I will hand over to Lucas who will just demonstrate the power of Navi we have at this discussion. Before before that, Gigi, quick thing. I know everybody is eager to look at the demo. We want to make sure that we have enough time for it. But a couple of very quick questions, probably, they maybe a one one or two word answer is what GenAI is Navi based out of? OpenAI, Gemini, etcetera? OpenAI. Right. OpenAI. Okay. How does an organization confirm that the data that they have is ready for AI? I think that's something which would be a great question for, Lucas to talk about when we does the demo. Maybe kind of cover that. So I think we we will, start the demo. Lucas, I'll stop sharing so that you can share your screen. Excellent. K. So let me make sure we have you on the Coupa platform. Okay. So, now let's just take a look at, the workflow that a user of the Coupa supply chain design and planning platform, would be using to leverage, Coupa Navi as part of their day to day activities. So I've logged into the platform. For those of you that haven't seen it, what we're focusing on is the heart of the technology, which we call supply chain modeler. This is where we're gonna take a a deeper dive into into Navi. So I clicked into that and very quick, navigation. What we have for the demonstration is a client, a user has gone through the, initial setup of the, initial digital baseline of the of their supply chain network. So we have the Coupa platform populated with the relevant data that Navi then uses for analysis. So one of the questions was, is my data ready for, AI? In the context that we have, in in front of us right now, you are populating relevant Coupa tables that Navi will then use for its interpretation, and analysis. So hopefully, that helps you understand a little bit more. The fact that you are working with Coupa will get you ready for for using GenAI. And what a what a user would go through in this example here so I have some data populated. We went through a few data tables. Just a quick overview. That user would then be creating what we call scenarios. So here we have the traditional type of melt design scenarios. Let's optimize for product flow. Let's optimize for lease and expansion, various strategic and tactical scenarios. Now that person then will go into our launchpad, select the scenarios, use our cloud, to solve all of all of those scenarios, and data will then populate it into our output tables. So this is kinda where I wanna start the shift into the Navi conversation. Historically, a user will have access to the detailed data tables. And one of the key challenges is, how do I make sense of this data? How do I interpret the results? And historically, you can send this out to Excel. You can look at the data and make sense of it. The other way that users make sense of complex data is through visualization technology. So I'm just opening up a visual example here of what's behind this particular model so you you have that context context. You know, I have a an an end to end supply chain model here. I have the dashboard that can start telling the story. This is now where I wanna open up for the GenAI to come in and be that third method, access to the data, the utilization capabilities, using the platform, using Coupa Software Navi to help me make sense of it all. So if I navigate to the top right hand side of the screen, you will see the icon for the Navi, AI assistant. And I have prepared a little bit of a story line so that we can understand how a user would be taking be taking advantage of this on an ongoing basis. There are two existing agents that have been made generally available to our clients. The first one is what we call the knowledge, base agent. So this agent is gonna be providing immediate answers on the latest product features, help them learn concepts of how to do models. Think about all of the challenges of a new user trying to get started with, you know, modeling an end to end supply chain. So this is a first example of where this technology comes in for that ramp up of new and existing users. The the question that I have as an example here would be, tell me about the latest features, the latest developments for network optimization. Hooper is always developing the technology. And as a user, I need to keep up with what's new. As opposed to having to go into a help file, reading through, you know, long detailed notes, I can just ask the platform. And as you see here, Navi will highlight for you what are some of the latest key features, of the platform. Not to go into too many details here, but what Navi's articulating is that we've, developed and enhanced our capability to do, invisibility diagnosis within network optimization. It's also gonna give you some information on, what levels of control this feature is gonna give you, how you can use it, and off you go. Right? So you're now more informed. You're having a conversation with the platform. Navigating to the second question that I have prepared here would be maybe a more sophisticated, still basic, but if you are new to supply chain design, if you are a VP of supply chain and you're trying to get your team to get up and going on this technology, These are the initial types of questions that they're gonna struggle with in the beginning. Right? But now it'll be much easier because the platform can talk to them. So what tables and fields do I need to populate to create a greenfield optimization model? So a greenfield optimization model is telling you what is the ideal location for, for your existing network, a very common exercise within the supply chain space. And as opposed to having to go through through detailed training, a user can simply ask, what do I need to do to accomplish this business problem? And as you see here, Navi will give you easy to follow instructions to create a greenfield optimization model. You need to populate the following key tables, customer's tables, demand table, and the products table. Right? They'll also give you context around what are some of the, fields that need to be looked at such as latitude and longitude. All of these things are historically done through word-of-mouth. You go ask a friend or you have to, you know, call somebody that can coach you through this or through significant trial and error. Right? Now you have the answer right away. For an existing user that's a little bit more sophisticated, and notice that my questions here, they're growing level of complexity. We start with the basic, and we get a little bit more sophisticated. So as an example, I'm an existing user, and I say, I currently source, products duty free from Mexico into The United States. However, now I need to model the potential impact of a new duty of 15%, whatever the percentage is, on products coming into The United States from Mexico. How can I model this behavior and give me step by step instructions? Now notice here that I am going beyond the basic of how do I do x y z. I am giving it a business problem that I want translated into a model. Right? So as a as an existing user, that's one of the challenges that we have to always overcome as existing users, which is how do I translate business requirements, business problems into technical language so that that I can build a model and solve it for my for my executives? So that's what Navi is helping us accomplish, in this example. So you notice the answer. To model to model the the impact of a new 50% duty on products sourced from Mexico into The United States, follow the following steps. Alright? So a little bit similar to what I showed you before, but the question was given a little bit more into a business type of situation that this system is able to interpret and and, have a conversation with you on what to on on what to do about it. Okay. So now real quick. Switching gears, we've talked about the knowledge management. Now I wanna talk to you a little bit about the network modeling agent. So this is a more sophisticated, more robust capability, which is helping users within network optimization, understand the results of their models, of their scenarios. This is a big deal. You have a complex supply chain, and then you have to make sense of it and communicate to your colleagues what is going on. So now we are enabling the platform through Coupa Software Navi to tell you what's going on so that you can be faster, more precise, more confident in in how you go about communicating results. In the past, most likely, this was done through what we see in the screen. Right? We have beautiful dashboards. We can filter through scenarios. We can drill down into different aspects of the system. Maybe I I need to go back into my data and send something to Excel and do a pivot table. But what if I can just ask the question? So this is where we're going back to Coupa Software. Right? So this is the modeling agent. Maybe we start with a a basic question such as such as, tell me, Navi, what is the total cost of the baseline scenario, of any scenario that I have in here? As opposed to looking into the dashboard, looking through the output tables, I can just ask it. Right? So maybe I'm in a meeting, and I don't know the result of a particular scenario as opposed to saying to my my director, my my manager, let me get back with you in a few hours or or after the meeting. Let's just ask Hooper right now. So So what we what we see here, we see the total cost of that particular scenario being, output into a summary view that I can easily read and understand. Additionally, there are artifacts that the system will create for you that allow you to drill down into more specifics on what's behind that result. So a very simple question here. The total cost for the baseline baseline scenario is 361. Excellent. Navi will give you, information around the assumptions that it's made in order to get to that result, what's included, what's not included, into that result. And a nice feature that I wanna highlight here is it's context aware. Right? So it has the ability to retain the context of what scenario we're talking about, what inform what is the context of our conversation thus far? So in this example, the system is asking, do you need a breakdown of this total cost by type, by site, by product? I asked for a summary view. Right? So the system is kinda leading you into a lower level of detail, which is truly fantastic. So my response to this was, yes. Break it down by cost type, and it generates the answer that you see below. So the total supply chain, cost for the baseline summary is x, and then it's gonna tell me what is the itemized view by type for for that particular cost. Right? So I can click into the artifact, look into the tables that it has generated. I can very easily see how that cost is, allocated across different cost components. And I love always reading the observations. Right? Because this is what technical users, call them modelers, call them analysts. This is a struggle that and I'll say this as a personal experience. We sometimes struggle translating the technical language into the business language in a way that others can consume. And I think the GenAI to Coupa Software is helping us get better at this. Alright. So these observations are things that I can now use to better communicate, with my peers. Another power, for example, that I want to give you, is the ability for it to create charts and graphical analysis off of your data. Within the Coupa context, this is both for output results as well as input tables. Very, very powerful. This is an example that a customer actually shared with us. This person wanted to create a distribution, chart of their customer orders, as well as a threshold on what is the ninety fifth percentile for customer orders. Now you can do this in many ways, but how beautiful is it that you can just ask the system to generate that for you? So this is the chart that it has generated. So this is a view of demand by customer. We see where the ninety fifth percentile is. If I wanted to print this off, I could do that. If I wanted to download this, I could do that as well. If I wanted to make this a little bit larger into a new window, I could do that as well. K? So giving you a little bit of the range of how you can go about using Coupa Software before analysis. The last example that we have right now, is the comparison. Just a quick example. Right? So, I have a scenario where we're doing a facility close, for for, whatever business, reason. A facility in my network has been closed. It's unavailable. And I want to understand what is the customer that is gonna suffer the biggest increase in service time because of this facility closure. Simple business question that would come up in a meeting. Right? A little bit more complicated to get that answer out of the model until now. Right? So think about the ability to type in that question, let the system run that for you, and you know that the customer, 0217, maybe that's the ship to location that, that you have in in your network, has experienced the largest increase in average service time. So it is comparing the baseline to my facility closure scenario. So it's going through that comparison and telling me that that particular customer has a a lead time increase of over twelve hours. And then as I have clicked here into, the artifact, you notice that I can lead the analysis, get more context. I have a quick table that I could use for sending out an email if if it's something that needs to be done quickly. You know, off I go. So the time savings are really tremendous, and, hopefully, this is just a little bit of a teaser of what's possible with Cupanavi in the time that we have together. I'd like to, pass it back to to my colleagues here for questions and final comments. Sure. Thank you. Thank you so much, Lucas. We actually have quite a few questions. We've we've been trying to answer those behind the scenes, but I think some of those questions, will merit an answer from you or Gigi as well. Kind of the typical question which was being asked from existing customer is how can we try out Navi? You know, some of them have, for example, cost to serve use cases. Others have, network optimization use cases. I think the the answer to that is yes. You can work with us, work, work with the the CBM of, from Coupa Software, and we'll be able to provide you some credits, to try out the functionality. JJ, do you have any other, things to add on top of it? Yeah. So all our existing, like, existing and new customers would be getting certain number of credits. I just work with the CVN, to experience the power of Navi. Cool. That's great. That's one more question about are there any limits to how many questions you can ask Navi in a single conversation? So, Navi works on credits. Essentially, you buy the credits, and each of those question which you ask, depending on the complexity in the depth you are looking from Navi to solve for, will be consuming credits. For example, knowledge management question which Lucas has demonstrated is pretty straightforward. Like, it will go into certain sort of dataset. It will mine through that and put it back. But so it will it will assume it will take certain number of credits, say, notice it numbers probably eight, nine, 10 credits. I'm just putting the number. But if you ask for a very complex reasoning question, why this versus that, and can you give me a graph and chart to make all of that, you gotta understand. Like, you know, it has to go to the multiple tables first, mine the data, stitch the thread, all of the data, and then get a chart and bar graph for you, and then come back with the existing escalation. So it's something like this would be more extensive or I would say, it is for Navi to come back. So there would the usage of credits would be higher. So, essentially, each prompt consumes some credits, but it depends on what your, what your what your specifically asking Navi to do. Yep. That's that's good. One other question which has come up is how does this change the implementation? You know, like, typically, when you do an implementation, you build a model and so on. How can you add Navi on top of your existing implementation? Nothing. Honestly, there's nothing. There is nothing for Navi. The moment you you do the the normal implementation, essentially, you have, the data ready for you to put in your data flows or any ETA workflows you have, push that into supply chain modeler. You can if you have Navi credits available with you, start on the Navi. Navi will automatically learn through your data, which you are already there in the platform, and start working on it. So for specifically setting up the Navi, there's nothing you have to do. Absolutely zero. Yeah. That's that's great news for the customers. One question, about the fact that Navi, Lucas, I think it's for you, mentioned that Navi would be able to give financial insights like PNL comparison. Will there be any specific screen to see those financial insights, or it'll happen within the chat of Navi? Think you're on mute. Yeah. As as of right now, it happens within the chat, that that you saw. It does create artifacts that you can download, you know, take a screenshot of. But as of right now, that is, that is done through the user interface that we demonstrated. Perhaps Gigi has additional context of future ways that the users can leverage it. Yep. I think, Lucas is perfect. So whatever your question you ask, it is being there is a summary of it and there's a detailed explanation, which Lucas is offering as an artifact. And an artifact is what you can save, even you can download, you can share. So that's the place. It's a ready record for you to just consume it directly rather than mining fluid. So that's the place it would be. The future of Navi actually goes into embedding with with UI. So, essentially, Navi will be able to directly take you to if something is available in the product itself, then it will take you there for more detailed explanation. So, essentially but right now, everything is readily available in the summarized format for you via artifacts. Great. Thank you. Thank you, Gigi. You know, there's still still a lot of questions. You know, we would definitely want to answer those questions. So, you know, we have an opportunity for you to engage with us. We have a poll here in front of you where wherein we are asking you if after the learnings of today's session, what next step would you be able to take or would you like to take as you con to continue to explore AI? You know, you could you could join add additional AI academy sessions dealing with procurement or finance. You could register for an upcoming AI live demo. We have one on November 18. You're welcome to share this recording with your team members, and then you can also contact us for a one on one, AI demo. And for those who feel like AI is not a priority, we really encourage you to look at how we can explore AI because I think that's becoming like, table stakes in organizations today. So if you could kindly let us know how you would like us to follow-up with you through this poll. Okay. So I think we, we're receiving a lot of good responses, to the poll, and I just wanted to also let you know that many of the concepts that we outlined in this particular, webinar, we have a white paper that we have created as well. And we'll be sending the white paper to all of the people who have, attended the webinar, today. That should really kind of talk about some of these concepts. So with this, I I really wanted to again thank, Ayush and Deloitte for setting the stage for us and talking about their experience with their customers because it brings a lot of, opportunity for companies to explore how they can gain value through AI. And thank you so much to Gigi and Lucas as well and to the team behind the scenes for making this happen and, of course, to the audience for asking these great questions and working with us. Thank you so much.