Thank You! Enjoy the Webinar

Speaker 1 (01:02):
Hi everybody. I’m Bob Bowman, editor in chief of Supply Chain Brain. Welcome to AI Wars, the Data Awakens presented by GAINS. Quick reminder, before we get going, there will be an audience question and answer session at the end of this presentation. Audience members are encouraged to submit their questions at any time during the presentation by clicking on that q and a icon at the bottom of your screen. So artificial intelligence today has become a business imperative, but for many supply chain leaders, the challenge isn’t whether to use AI, it’s how to use it effectively. The promise of AI in the supply chain has never been louder or more misunderstood. While many organizations continue to debate the definition of real, AI leaders are already using machine learning optimization and agentic systems to improve planning agility, forecasting accuracy, and decision automation. In this second installment of our AI webinar series, we move from theory to action.

(02:11):
We’ll explore practical real world use cases like intelligent network design, anomaly detection, automated vehicle routing, and sequential year planning, and will help you identify AI ready opportunities, avoid common traps, and understand how AI can reshape your supply chain, not as a one-off project, but as a component of a dynamic decision framework. So let’s review what we’re going to talk about today. I’m going to start out by asking how do we see AI today? Then where and how should AI be applied effectively? Where is AI already delivering real results? And then we’ll move into our thought leadership panel in which I have the privilege to ask our speakers of today some questions. And then we’ll wrap up with the previously mentioned audience question and answer session and some final takeaways. So with that, I want to introduce our speakers for today. Matt Morton is Senior Director of Engineering at GAINSystems, to which he brings more than 15 years of your expertise in advanced analytics and machine learning.

(03:20):
Matt is the technical co-founder of supply chain strategy company 3TO, which was acquired by GAINS in 2023. At GAINS, he leads the advanced research, data science, and operations research teams. Antonio Nanni is a data scientist and AI engineer at GAINS. He holds advanced degrees in philosophy, statistics, and sociology, and has been researching the inner workings of generative AI since 2014. Antonio’s contributions include the development of the lead time prediction machine learning routine. Jeff Metersky is the vice president of solution strategy at GAINS. He has more than 35 years of experience in supply chain management. Jeff co-founded ChainAlytics and has held leadership roles at top firms including i2 Technologies, Llamasoft and Optilogic. So with that, I want to pass it over to Jeff for the presentation. Jeff Metersky, take it away.

Speaker 2 (04:22):
Great, thanks Bob. Pleasure to be with you today. And your listeners really excited to talk a little bit more after our first installment about what AI is all about. And so I’m going to start with, it’s been interesting to me as I’ve been in this space for a long time and I’ve seen the uptick that’s happening, a significant uptick in interest levels of AI. And there’s a variety of different surveys that are out there. I just selected from one of the many, one from Thomson and Reuters that was a 2025 C-suite survey of over around 200 senior level executives from various countries around the world. And the reason that I kind of honed in on this one is we’ve just been seeing from not only our current customers, but prospects in the market, significant levels of requests from boards to say, go do something with AI.

(05:17):
And what they’re going to go do with it is not necessarily known, but it’s imperative that you have to do something. And what I found really interesting about this survey is when you look at the top 10 priorities and look at the top three being digital transformation, improving operational efficiency, and then exploring potential in implementing AI, just take a glance really quickly of the significant change that’s happened from 2024 to what’s being predicted. That’s the black bar to 2026 with a red bar. But at the end of the day, it’s not just implementing AI or exploring AI for the sake of a new technology. These executives are still focused on outcomes. And so it’s important for us to remember and consider when we select a technology or a process improvement or any initiative, what is the actual outcome that we’re driving against? And these are not really significantly changing, right?

(06:12):
The top two are revenue and profit, and then they kind of fall back down. But you don’t see a success measure of saying, how many AI implementations have I actually done? So I want to make sure that I’m always tying the use of a technology to an outcome. And for us, when we think about how we see AI, so this is a GAINS perspective and how we believe that AI can actually be incorporated into our tool set. So we think about AI going down from a starting point of an umbrella of two major buckets of machine learning and natural language processing. And then underneath machine learning, you’re diving deeper and deeper, pun intended into deep learning and then gen ai. But what you’ll notice on here is two other complimentary areas that we do not call ai. There’s a lot of AI washing that’s going on out there.

(07:09):
We can say that every tool under the sun is AI oriented. We think that there is a group of technologies and approaches, and I’m going to talk about them in a minute, that is data science centric, that’s operations research centric that are complimentary to the AI technologies, then could all be orchestrated through ag agentic ai. So AI itself is a tool. It’s not a strategy. When we think about what’s available for us as supply chain leaders to make better decisions, we want to pick the appropriate tool for the decision and the outcome at hand. And so while it’s easy to say, let’s pick up that AI hammer and just bang it at everything that we can possibly find and we’re going to improve it, we don’t see that that’s the way to improve outcomes. We believe that AI is a way to augment and not replace human judgment.

(08:08):
That at this stage of the game, while there’s great promises of the future and of autonomous supply chains and self-guided supply chains, we’re not there yet. We continue to want to explore and evaluate and test. But right now human judgment is imperative to get real value out of the current state of where we’re at. We want to focus on decisions. What are the decisions that you’re trying to make and support and the outcome that you have in mind? And then let’s pick the right technique, right? AI is just one tool that we can leverage, right? It’s not the only tool. And it’s actually a great way to compliment and use other tools together in harmony to improve our outcomes and get a better decision. We absolutely want to have an understanding of data readiness and quality. It’s been said by lots of folks, and we’re not any different when we deal with technology, when we deal with decision support, decision-making tools, we need to have quality data.

(09:03):
We don’t have to have perfect data. And in fact, AI is a great way, and I’m going to talk about it in a minute, of how we could use various AI techniques to improve the quality of the data that we’re using to feed not only other AI applications, but our operations research and data science tools. And then finally, what is the architecture that we’re going to need to really leverage this, right? We believe in a lakehouse enabled composable and scalable solution or bits and pieces in a service oriented architecture. We can be plugging in, we can be enhancing individual decisions, but we don’t have to necessarily bury this inside of our large scale monolithic solutions. With that, the way that we look at this, and this is a slide that we represents, the tools that are in our toolbox today that we’re all really familiar with, or many supply chain leaders are very familiar with how to use these tools to support a spectrum of decision-making.

(10:00):
This is not ai, but it’s important for us to understand where are we going to apply our different techniques from descriptive with what happened to diagnostic, why will it happen predictive? What can happen prescriptive, what actions should we take? And then really that last layer that’s come on in the last several years is cognitive, right? And you’ll see that there’s a variety of different techniques that exist today that we classify as data science or data-centric types of techniques. And they’re incredibly valuable. There’s a set of decision centric types of techniques that fall under a category of operations research and then a genetic ai. And as I move across the spectrums of my analytics portfolio, I don’t want to forget to take advantage of these. I’m not just going to throw them away because I now have that AI hammer doesn’t mean that I’m going to look to replace all of this.

(10:53):
I’m going to see are there places where AI can be applied to improve the quality of what I’m doing? And so when we think about that, and this is inside of our framework, excuse me, of what we call decision engineering and orchestration. We see multiple areas where we can be applying AI in the data cleansing world across this entire spectrum from descriptive to prescriptive anomaly detection and outliers. Can we improve the quality of our data? Can we make automatic corrections like removing duplicates? Or if I missing data, can I infer or impute holes in my data that I can plug in as I’m trying to do data integrations across multiple systems? Can I do a better job of harmonizing that data? And then a big part of what I can do in AI that I can’t do with these other techniques is transform and an organized unstructured data.

(11:44):
How do I take sentiment information? How do I take things that are inside of documents? How do I take social media inputs from a data visualization? And can I actually decide, instead of me having to be the expert to figure out do I need a bar chart or I need a scatter diagram or whatever it may be, can I auto generate the most suitable charts and dashboards? Can I get a better idea of my key trends in relationships? And then how do I use natural language, right? We are moving into environment, as many of you know you’re familiar with using things like chat, GPT, can I actually speak, right? Can I type in natural language? Can I ask my questions in such a manner that I don’t have to actually programmatically do the work? And then finally, in the level of visual assistance, where can I go for self-service support, right?

(12:35):
So this is your typical type of real-time guidance and chat bots, but all the way up to how to instructions. Can I guide the less informed individual or the person that’s not an expert how to do something? Can I actually do automatic model generations and reviewing my results and drawing conclusions? And then in the world of operations research where we heavily rely on optimization models, can I detect and resolve modeling feasibilities? But where we want to spend the rest of the conversation, and I’m going to turn it over to Matt in a second, is where we have been focusing in addition to those in specific supply chain centric opportunities to apply ai. And that’s in the predictive area of what can happen around forecasting demand and lead times. And we’re going to talk about order creation and management. So when I’m making decisions of what I need to buy when I’m replenishing my inventory, what suppliers do I select, how big should my orders be? And then managing that on an ongoing basis. So we’re going to focus on moving that over to the conversation. And with that, I’d like to introduce Matt Morton, who’s going to take it from here.

Speaker 3 (13:50):
Thanks, Jeff. The first topic that I want to cover today is demand prediction and forecasting. So most forecasting today is fundamentally backward looking. It relies on historical data, and this creates a blind spot where we’re focusing on a lagging indicator. It tells us what happened yesterday and not necessarily what’s coming tomorrow. Think about what this means operationally. You get no early warning for economic shifts, market shocks, or sudden changes in demand. By the time these signals show up, you’re already looking behind this forces planners into a state of reaction. They’re spending their time firefighting last minute exceptions rather than proactively shaping supply chain strategies. When all your time is spent researching exceptions and making adjustments to them, it takes away from strategic activities that have a longer term impact. This creates a hidden drag on operations. It reduces slows decision making and ultimately impacts your ability to serve customers effectively.

(14:59):
And so where can AI help? It’s in combining the backward looking traditional time series methodologies with forward looking metrics and advanced learning. The architecture that we’re showing here is pulling in both internal operational data and external market signals. We’re talking about financial market data, stuff from the Federal Reserve and also internal material level demand patterns and engineered features. The key insight is that you’ve got continuous learning from data, both what’s happening in your business and what’s happening across the broader economy. Rather than one model, you’re building bespoke prediction engines for different customer segments, product categories and geographic regions. The combination of internal and external signals creates higher forecast accuracy and deeper business insights. It transforms your planning from a black box exercise into a strategic capability.

(15:54):
So how does this affect outcomes? So here are some real world results that we’ve noticed. There are three companies here across different sizes, different industries. We’ve got an industrial automation manufacturer, a jewelry manufacturer, an electrical distributor. What we’ve noticed is a consistent 20 to 30% reduction in forecast error for all these different companies. This highlights the ability of the process that we’ve outlined to directly drive inventory turns and service level performance for supply chain leaders. It represents a shift in how you can approach inventory planning. When you trust your demand signal more, you can operate with lower safety stock while maintaining or improving service levels.

(16:43):
The next section I want to cover is on lead time prediction. This is where it’s arguably even more impactful than in demand prediction. And the reason is simple. One day of lead time is equal to one day of supply. Lead time is too often treated as just a static number. It’s set once it’s set based on a variety of different things. It’s set based on history. But the fact of the matter is, if you overstate lead time by just two days, you’re carrying two extra days of inventory across your entire network. For large organizations, that means millions in excess stock and tied up working capital. If you understate lead time, it introduces massive risk to your supply chain from stockouts and service failures making lead time predictable and precise unlock significant value for large organizations. Potentially you’re in the range of hundreds of million dollars in working capital.

(17:45):
So what are the challenges? The traditional approach to lead time management has a fundamental flaw. It assumes the future will look like the past. Most systems calculate lead time based on historical data input from suppliers. But what happens when you onboard a new supplier? What happens when you launch a new product sourced from a new region? You’re flying blind until you accumulate enough data to make it worthwhile. In today’s environment, with a huge base of supplier diversification, nearshoring trends and supply chain evolutions, this backward looking approach is inadequate. Advanced AI solves this by moving beyond historical order data to analyze underlying drivers of lead time variability. We’re looking at material characteristics, supplier attributes, transportation modes, seasonal patterns, and macroeconomic indicators. The goal is to understand lead time at a causal level, not just a statistical one. This gives us the ability to estimate lead times for new supplier product combinations. From day one, you’re leveraging patterns learned from similar materials, similar suppliers, similar trade lanes, similar vendors. The explainability component that we’re introducing here is also crucial for adoption and ongoing maintenance. You can see exactly what factors are driving lead time prediction, supplier performance, material complexity, logistic constraints. Instead of just lead time being a planning assumption, it becomes a strategic insight.

(19:14):
So we’ve got one case study to share with you In this electrical distributor faced a lot of the same challenges that many people did during the supply chain. Disruptions of recent years, lead times increased by on average 80%, some items increasing as much as 300%. The real problem wasn’t just longer lead times, though it was unpredictability. When you can’t trust your lead time assumptions, you have to increase your safety stock. You have to handle that risk somehow. For a company of this size, it was costing around seven and a half million dollars per day of lead time. That’s a strategic problem. The machine learning approach that we implemented gave them the ability to predict lead time changes before they impacted operations. It was early warning rather than after the fact adjustment.

(20:05):
I think the results really speak for themselves. It really shows what’s possible when you move from reactive to predictive lead time management. We saw a 65% improvement in lead time accuracy. We saw an 18% reduction in loss sales and a $21 million reduction in inventory. The ROI is particularly noteworthy, 967% in 1.3 months. This isn’t a multi-year transformation project. It’s immediate payback and immediate impact on your business. The last area I want to cover today is in order decision automation. It is one of the most resource intensive activities that a supply chain team can be taking across the organization.

(20:57):
So I would say most organizations still rely on a heavy manual process for processing supply chain orders. Human buyers are reviewing hundreds to thousands of orders weekly. The traditional approach will use custom business logic, custom business rules, minimum order quantities, working within budgets, trying to hit minimum order targets. A lot of times these rules though are static. They can become outdated and stale quickly. They don’t necessarily capture the nuanced decision making that experienced buyers rely on. So when your most experienced supply chain professionals are spending a significant amount of time on routine decisions rather than strategic activities, it doesn’t scale and it’s, it’s not giving you your best outcome.

(21:46):
And so what we’ve outlined here is an architecture for incorporating AI into enabling automated order and purchase order automation. And so we’ve got a few different aspects here. We’ve got a few different machine learning models that are predicting whether pos are good or not, at both a line level and a larger PO level. We’ve got a multi constraint order builder that sits in the middle of the process where you’re able to optimize across a number of different things. You’re ordered, not just minimum order targets from vendors, but things like how much can fit into a truck, how much can fit into a container, as many different constraints as exist in your business, it can handle. And really the key component of all of this is the last one here where you’ve got our AI process that’s learning from the best buyers. And so we’re not just trying to emulate the average buyer but your best buyer. And this improves outcomes across the board. And so what do we see on results? So I think it’s, it speaks for itself. We see less than 5% of order lines requiring manual intervention, less than 20% of orders needing human review. But the strategic impact is even more important. Your buyers can shift from administrative tasks to value added activities, supplier relationship management, strategic sourcing, strategic supply chain optimization. It’s a fundamental shift in how teams allocate their time and expertise.

Speaker 4 (23:31):
That concludes

Speaker 1 (23:32):
F. Thank you very much, Matt. Let me bring Antonio and Jeff back in here at this point and greeting all of you. Thank you everybody for being here today. This now is the panel discussion portion of our presentation, at which point, as I said before, I have the privilege of asking these experts some questions about this subject. And then when we’re done with that, we’ll move on to the audience question and answer session. And even as we are talking now, audience members are encouraged to continue to submit your questions by clicking on that q and a icon at the bottom of your screen. And we’ll get to as many of your questions today as we can. Time permitting. So lemme start with you Jeff. One of the points in the presentation so far is that AI isn’t just about adopting new technology for technology sake. It is about supporting decisions in order to drive better in outcomes. So how do you see this shift in focus influencing how supply chain leaders are evaluating and adopting AI and balancing board requests?

Speaker 2 (24:33):
Well, it’s a great question, Bob. I mean, I would start with the answer of I don’t see enough of it do influencing it. I think we’re taking a big shotgun approach. We’ve got that as I was mentioning in the presentation, we’ve got a hammer and we’re looking for that solution at the end of the day, where can I go out and apply it? And I’m going to try to find as many places as possible where we have found success. Most recently in this, and Matt has shared some of those examples of where we decided to focus our attention is on truly understanding where is your biggest pain point? Like look at border states. I mean if they looked at their lead times and the inability to predict their lead times, that was a major frustration for them that had ripple effects everywhere else. And so they didn’t come to us with a, can you use AI to solve this problem?

(25:22):
They came to us with a challenge that they had. We’ve looked through the plethora of alternative techniques that we have at our disposal and came to the conclusion that to move the outcome of getting a better lead time that influences inventory, that influences lost sales, that we can adopt a machine learning technique, but we’re not throwing away everything else. So for me, what I think is really, really important is this is not a new way of doing business in terms of what are my decisions, what are the techniques available? We just have another tool in the bag and I think that too many companies are doing this shotgun up. I must find a way to use AI as opposed to where are my biggest pain points? And then do one of the many AI techniques really apply and do they make sense

Speaker 1 (26:06):
Cutting through the hype because there’s so

Speaker 2 (26:07):
Much, yeah, cutting through so much hype, right? Yeah, being pragmatic about it. Absolutely

Speaker 1 (26:11):
Perfect. Yeah. So Antonio, there was a brief mention of ag agentic ai, but it was kind of glossed over a little bit. What is ag agentic AI actually in practical terms and how close are we to this concept of agentic agents being mainstream and supply chain? And then finally, what are some realistic use cases today?

Speaker 5 (26:34):
Well, thank you for the great question, Bob. We’ll start with the easy questions. I love that. This is a very hard one. So me start with the definition of azen, KI. That’s the easy part. So we can define azen, KI as a large language model such as JG, PT and all the successors plus tools. And in this context, tools are interfaces between the world and the model. So for example, a tool can be something that led the model query your database or trigger a PO approval. So what can we do when we have tools connected to large language models? This is a multi-billion dollar question, honestly, everyone is trying to figure that out. And it’s very hard right now to understand where this technology will ultimately land, at least in this current technology cycle. So I can tell you what we are doing at GAINS right now to prepare ourself for this.

(27:40):
And I really want to echo what Jeff and Matt said that this is a tool is not the only solution to all the problems and the shotgun approaches fundamentally flaw. So what we are doing right now is to drop every service we have as a tool for gent K to use. So the lead time prediction, service demand prediction, PO approval, and many other services that we are creating against will be wrapped up as tool for Gent K to interact with. And the idea is that a gent KI will then drive the user by knowing the in and out of those tool. And the ultimate goal is always to help the user make better decision. The point here I want really to come across to capture the AI as you said, is that ai, sorry, Azen AI by itself is not the solution to all the problem. It needs to be built on top of other solutions that already deliver value.

Speaker 1 (28:38):
Yeah, well, excuse me, the word that keeps popping up in any discussion of ag agentic AI is autonomous. Now you earlier referred to prescriptive AI as what action should we take emphasis on the should implying that now we’re turning it over to the human to actually take the action Is ag AI that next step forward where it is taking the action and we are taking our hands off the wheel,

Speaker 5 (29:05):
We’re not still there. I can refer to some, for example, the last blog by philanthropic where they talk about the approach to ai. It’s really important to be cautious, especially when AI can make decision that really drive your business in both direction, both for the good or for the bad. So we’re not still in a place where it’s Zen AI can take decision for itself. We’re in a place where a more proof system like PO automation can help Zen K show you what decision you should be making. It’s more of a assistant for the user than a substitute for the user, at least at this point in time.

Speaker 2 (29:50):
And I would just add on Bob, I mean I think we continue to look and believe that it may get there at one point and we want to see things are rapidly changing. There’s technologies that our team is using right now that six, nine months ago weren’t even available to the general public or general developers. So this space is moving rather rapidly, but we do want to be cautious as we move to that self-directing supply chain autonomous, that might be a vision out there where can we make decisions that move into it’s appropriate to take the human out of the loop and there’s going to be some of those where it makes sense and the risk is not huge. Some of this we have to understand in any decision we make, what’s the risk of being wrong? And so those places to start testing and applying, this is where the risk is not great that if something wasn’t done correctly, I’m still okay. I still can recover. I still have enough resiliency as opposed to putting this on the, I’m going to run my plant 24 7 autonomously and if something goes down, I’m in deep. You know what? So I think it’s just a matter of let’s be, we are cautiously optimistic. We are not saying no, we’re not saying that that can’t be a future state, but we are not there yet, at least in the areas where we apply and think about supporting our customers.

Speaker 1 (31:08):
Thanks for clarifying that. Matt. We’ve seen some strong numbers here on lead time improvement in this presentation, lead time prediction being one of the first applications I guess of ai. In your estimation, what made that such a strong candidate for AI in the first place and what is required behind the scenes to make that use case successful?

Speaker 3 (31:30):
Yeah, that’s a great question. There’s a few different ways to answer it. From a technical perspective, lead time is often affected and driven by things that are easy for an AI or a machine learning algorithm to detect and recognize patterns for. But it’s a hard process for humans to manage proactively. It’s one of those things where if you’ve got too much lead time set, it just kind of flies under the radar business as usual. Nothing really rises to the top from this is an urgent thing that I need to go look at. And so it’s a great candidate from that perspective. I think from a more human perspective, there’s not a lot of people who are focused on this and we saw the pain points in our customers and so the results really speak to themselves for the ROI that we’re seeing from our customer base. And it’s a great story.

Speaker 1 (32:26):
Yeah, certainly COVID-19 certainly stretched out lead time to such an extent that companies had to wake up to this notion that something had to be done about getting a handle on it, I’m assuming. Yeah. Okay. Well, Matt, let me stick with you for a moment. We are often hearing that AI is going to replace traditional supply chain techniques. You guys have done a great job today of explaining how that’s not necessarily the case. How do you determine where to apply AI to gain the most value? And for that matter, where do you see them actually replacing these techniques?

Speaker 3 (33:00):
Yeah, I think that’s a constantly evolving sphere. And so for us it’s really about outcomes. We’re not looking to do this broadband across the board, but really diving deep into each individual place where we could possibly use AI and understanding is this going to move the needle from a business perspective, but also understanding that that’s a moving target though the pace that AI is improving means that you have to constantly be on top of that. And so we’re investing the time and resources and expertise to make sure that as the capabilities change, we’re there to meet it too.

Speaker 2 (33:36):
I think a big part of it as well, Bob, is there’s this notion of a return on investment. I think about the ideas of I have multiple different ways that I can get from point A to B. I can walk, I can take my bike, I can drive, but at the end of the day, there’s all different constraints that are literally put on in those situations. And I have different goals and objectives and I can think back in my career where I’ve done a lot of work on optimization, like in that operations research area of applying operations research, and there was problems 30 years ago that were intractable to be solved by operations research types of techniques. They were ideal. But if I was trying to get an optimal answer, respecting constraints that was minimizing costs and I needed it for a real world application with a decision to be made in a minute, I can’t get there.

(34:30):
And so I had to take a different technique at that set of time, some heuristic based type of methodology. And then as algorithms is proved as computer speed improved, we can keep on coming back and saying, I really would’ve liked to use that optimization technique. It would be better. Is it appropriate now? And I think that’s where we are in some cases with some of the various AI techniques, right? We need to keep on trading off how fast am I trying to make a decision? How much money do I have to make the decision? What’s the quality of the decision I make? And those are criteria that we start thinking about when we’re looking at the appropriate tools and techniques to bring to bear where does it fit and where does it not fit? But it will keep on changing. I think it’s going to keep on changing just like it did on where optimization techniques get applied to supply chain problems. As we keep on improving horsepower efficiency, algorithmic efficiency, we can start bringing them back. We just want to always be looking at what’s the best we can use for our solutions today, but never lose sight of the fact that that could be an ever-changing environment that’s not controlled by GAINS and it’s not controlled by any of us. It’s controlled by multiple players that are out there that are testing and pushing the envelope right on moving technology forward. But we like to be pragmatic about applying that to make businesses better.

Speaker 1 (35:44):
You mentioned the data and data science kind of exists outside of ai, and I’m wondering if back in the day when we were unable to make those calculations, we didn’t have the data access to data that we have today. So that

Speaker 2 (35:56):
A hundred percent

Speaker 1 (35:57):
Certainly amount of data, quality of data, access to data powers, this whole thing, it’s not necessarily within an AI subset. Antonio, I just wonder, I’d like to get your take on a question I just asked these guys about where to, how do you know where AI is going to give you the most value?

Speaker 5 (36:18):
I will give you a more technical perspective here since Jeff and Matt already spoke perfectly about the business. Nothing to add there. So I will say the two keywords here are adaptability and efficiency. And Matt already hinted to that in his own presentation. So with adaptability, we mean the fact that it’s common experience in what world do we live in right now, it’s a word where variability has gone up a lot. Things change very fast and radically, I would say month by month if not week by week. And so your system cannot rely on the previous lead time you observed to calculate the next lead time because that’s already too late, your order is already late. Once you do that, and efficiency again, Matt, hint to that in the presentation is the fact that we can use lateral learning. So for example, the demand for the historical demand for item A, B, C can actually inform the predicted demand for item X that you’re launching right now, or the same will go with lead time or PO approval. So from a data perspective here, I’m really talking technically you need to be adaptable and efficient because the world is changing a lot. You have new items all the time, and that is where our value really is.

Speaker 1 (37:54):
Okay, so Antonio, I’ll sticking in with you for a moment. I’ve heard you discuss composability in previous webinars and content, but I never heard you mention a Lake Lakehouse architecture before. What’s going on? Are we changing movie references here all of a sudden? What is this and why is it necessary to leverage ai?

Speaker 5 (38:14):
Okay, I’ll confuse you right now. That’s a great question. So there are three main architecture when it comes to data architecture. So is the data warehouse, the data lake and the data lake house, and everyone wants a lake house, obviously. No, I will not go into the details, the differences between all three of these architectures, they are painful, I’ll say that. But I’ll talk about the consequences of adopting composability and data lakehouse architecture. So what do we gain from this architecture? First of all, let me go back to one of the example that Matt gave with a jewelry customer. As Matt mentioned, we predicted demand for them. And an interesting request they had is that they wanted to incorporate exogenous data. They wanted to incorporate the publicly available price of gold future to predict demand for their jeweler. And what we did was simply, well, simple enough, we ingest the data in the data lake, data lake collects all sorts of data, both your classic SQL table as well as unstructured data and was very easy to incorporate this in the prediction we were making.

(39:41):
Another great characteristic of the data lake is that since you have all your data there spinning up service as a plugin, as Jeff has in its own slide, it’s very easy. The time to spin up a service is measured in terms of week, not in terms of months of year. We are really talking about an implementation time. That is usually some weeks. I don’t want to be too precise because obviously there is variability there, but that is usually the kind of measurement we have in our mind. So it really allows us to have this composable plug and play kind of solution that is very important, a great advantage for our customer.

Speaker 1 (40:31):
Well, I’ve been enjoying the privilege of hogging the questions here, but I think it’s time to bring the audience into this. We do have some audience questions already out here that we’d love to answer. And even as we are answering the ones that have already come in audience, please do continue to submit your questions by clicking on the q and a icon at the bottom of your screen and we will of course get to as many of those questions as we can. Time permitting. Let’s start with this one. Questioner says it’s impressive that you’re processing 80% of orders automatically. Where are you making the decisions? How are they implemented in the ERP enterprise resource planning application? Matt, why don’t you take that one?

Speaker 3 (41:12):
Yeah, sure thing. Yeah, it’s a great question. In the GAINS planning system that we have today, there’s already a mechanism for recommendations to be generated and customers can set up rules for how those are approved or needing review or rejected. And so the interest or in the GAINS perspective of building composable architecture for the different processes that we do, this automation process kind of sits in the middle of that where it sits an enhancement to that process and we’re able to work and give our customers levers to see the confidence that our own system has in the POs that it’s recommending through this new AI driven process and to increase that threshold for what gets automatically approved or not.

Speaker 1 (42:04):
Okay. We talked a little bit about this, but I think this questioner wants to hear more. Do you see AI replacing roles in supply chain or is human experience still going to be valuable? Obviously nobody knows for sure, but Antonio would love to hear your answer to this question.

Speaker 5 (42:23):
Yeah, I feel very strongly about this, and I should say before I even start that this is a perspective from August, 2025. We all know this space really evolves very fast and this meeting will be recorded. So I just want to give this as a context for future viewers of this webinar. So I really feel that AI is a tool and it’s a way to expand our capacity and honestly our mind. But this sounds very fancy and it is in a way, but I do want to take a step back and really talk about other ways in which our mind is expanded right here, right now. And it’s been for the last millennia basically. So once upon a time, if you wanted to hear the Odyssey say you had to find someone that knew the Odyssey by heart and they will play that to you in front of you in your court probably once upon a time, and we were talking the middle aged in the western world, if you wanted to do a multiplication, say 35 by 17, you would’ve gone to a monastery, given them your multiplication, and then they would’ve done that for you.

(43:39):
It would’ve taken them roughly a week, and obviously you would have to pay the Arab numbers. There is what really changed the game. Multiplying Roman numerals is very, very hard. Multiplying Arab numerals is something that we do in elementary school, and so this kind of tools really expand our mind. I see AI doing exactly the same thing. We see that in coding right now, where already we start to consider some tasks too menial, too easy to be done by hand, by ourselves, and we will go there probably in supply chain for example. PO automation may become a very, very standard tool. Eventually we cannot say that is not happening. But the idea of substituting, in my opinion, is a wrong framework to see this. I really think about in terms of expanding our capability and honestly also freeing our capability. What else can we do if we’re not doing this menial job?

Speaker 1 (44:46):
Okay, thanks for that. Questioner says, what new exogenous data sources fancy word for external data sources are going to become most valuable to enhance AI in the near future? Jeff, why don’t you have a go at that one?

Speaker 2 (45:03):
Sure. I think it’s a great question and it’s really broad. I don’t know if it’s enhancing ai. And again, kind of with the theme of what we’re doing, I think how can our decisions be better enhanced by using exogenous data? And I think it will depend on the various areas. If I’m trying to improve my demand forecasting area, then I want to try to take in things like weather data. I want to try to take in social media sentiment data. If I think about as a specific area, it’s very difficult if not impossible, you can’t use traditional time series techniques on unstructured data. But if I can actually take information from what’s happening in consumer chats, what’s happening in social media, and bringing that in as an indication that there’s a connection between an uptick or a downtick in my demand based on the volume of that activity, that’s an exogenous data source.

(45:59):
That’s not just about my orders that have been placed on me or my point of sale data. If I go to the world of trying to do a better job of understanding disruptions and when disruptions may actually occur, can I be looking at financial information about my suppliers? Can I be looking if there’s an uptick in bankruptcy that’s happening based on certain characteristics? So there’s information that allows me from an exogenous perspective to better improve the quality of data. Sometimes that feeds my other systems and applications. That’s not just AI by itself. And so I don’t know so much. It’s a great question. I haven’t given a lot of thought to what new sources. I just think there’s a lot of existing sources that we haven’t thought about how to tap into enhance this decision-making process. Not that it will make AI in and of itself more valuable, but there’s data sources that are out there that we’ve just never been able to use because we didn’t have these techniques. And so I might alter the question that they’re not new data sources. They’re existing data sources that we now can find ways to use to compliment our other technologies to make better decisions.

Speaker 5 (47:07):
Well, this question if I may add something.

Speaker 1 (47:08):
Go ahead,

Speaker 2 (47:08):
Please,

Speaker 1 (47:08):
Antonio.

Speaker 5 (47:11):
This really is a place where the data lake architecture shines. To go back to the previous question because another characteristic of the data lake architecture is that you have structured and unstructured data living alongside each other. So you may think, and here I’m just throwing out an example, maybe it’s not even the best you can find, but you may think, for example, of a workflow where you upload unstructured data source such as, for example, the PDF of a contract with a supplier, and automatically AI system extract information out of your unstructured data and for example, change your structured data at the SQL table, the price that comes from the new supplier contract in your supply table. So it really allows us to extract information from unstructured data sources and seamlessly integrate with structured data sources that then are the place where machine learning really shines

Speaker 1 (48:21):
Expanding on the thing about data. We’ve talked about data availability and quality, and they always say garbage in, garbage out. This questioner wants to know how perfect does my data need to be to start leveraging ai, Jeff,

Speaker 2 (48:37):
What doesn’t need to be perfect? And I think one of the things that I get excited about is applying these techniques to improve the quality of the data that I have. It’s one of the challenges that we always have. Data does not need to be perfect, but you do need to understand what you do have. I think too many times we go for that, and I love the phrase, and we’ve been using it forever, Bob, you right? Garbage in, garbage out. But it’s also what is the fidelity of that data? So I maybe I may not have perfect data, but if I understand what the variability or uncertainty is in that data source and how it might be driving my answer, I could still use it and I could still start somewhere with it. So many of us don’t make decisions around perfect data, and this is where my personal perspective is that data quality has a direct correlation to the autonomous supply chain and the cognitive decision making there.

(49:30):
If I do not have oversight, if I do not have a way of looking at the recommendations that have been suggested, then I’m going to have a higher degree of requirement for good quality data that’s going in. The more that I can actually look at the answers that are being prescribed and decide what I want to do, the less I need to have the higher quality data. So I think there’s that balancing act and there’s always ways to get started. We talk about, I’ve been talking about this for a long time. There’s a difference between I have to make an execution decision today about, let’s just say in the transportation world of a carrier that I’m going to go select, and that carrier needs to know exactly what my order is, exactly where it’s picking up from, what time, what’s the destination. If I’m making a strategic decision about the mode of transportation that I may want to use in my network in the future, I don’t need that level of precision because I’m not executing it today.

(50:24):
I can play more scenarios. I can play more what ifs. I can understand where the trade-offs may be. And I think it’s the same thing here, right? The closer I am to operational execution and the closer I am to being in an autonomous state, my data needs to have a higher level of perfection. The further away I am into tactical and strategic level decision making. I don’t need the same high quality of data as a starting point. We still want to aspire to that, but I think that’s part of how we can frame up the differences of where will this data be applied when we’re using ai?

Speaker 1 (50:57):
Okay, the questioner wants to know, do you need to be a GAINS customer to take advantage of the AI solutions we’ve talked about today? Matt, what’s your take on that?

Speaker 3 (51:08):
Yeah, the answer is no. We are building these solutions and these processes to be agnostic to the GAINS planning, existing, existing customer base and planning. And so you’ll be able to take advantage of it and the solutions that we offer to solve business problems from anywhere.

Speaker 1 (51:31):
Alright, well, hey, this has been a fantastic eyeopening presentation for me and I’m sure for our audience as well. Unfortunately, we are just about out time. We have time for one final question. I’m going to direct it to each of you. I’m going to start with Antonio and here it is. What’s the one thing that supply chain leaders should walk away remembering about how AI fits into their business and what’s the first thing they should do? Antonio,

Speaker 5 (52:00):
We live in a new world. Adaptability is a key word here. If your system is not adapting to the new conditions, then it might be automation, but it’s not AI and it’s not driving better decisions.

Speaker 1 (52:17):
Matt, what do you say?

Speaker 3 (52:20):
Yeah, I would say AI is not the right tool for every situation. It’s a rapidly improving area. It’s becoming more and more applicable in different areas all the time. But we’re not trying to chase ai. We’re trying to chase better decisions and then choose the technique that fits the situation.

Speaker 1 (52:38):
Jeff, last word to

Speaker 2 (52:39):
You. Well, Matt stole my answer, but I guess I would compliment it with, so it is a tool, not a strategy. Let’s make sure that we’re not throwing away all the other wonderful techniques that we’ve been using and find a way to compliment them. But in terms of what to do, what is that area where you are having the greatest challenge, where you have a KPI performance metric that just you can’t seem to get control on and see whether or not AI is applicable to be applied into that area?

Speaker 1 (53:10):
Well, fantastic. Thank you so much guys for helping us to cut through the hype about ai. At the same time, talk about the real value that AI is delivering today to the supply chain. It’s been a great presentation and audience, thank you so much your participation and your great questions as well. We have for you a complimentary white paper, ag agentic, ai we talked about that meets DEO orchestrating the future of supply chains. There is a QR code for you to take a shot of to access that, but if you don’t have time to do it before the end of this presentation, then no worry about it. It’ll be sent to all of you at the conclusion, which I have to say unfortunately is now. Thanks again everybody. It’s been great. Everyone have a great day.

Speaker 2 (53:58):
Thanks, Bob. Thank you.