Jerry Holbus 0:00:00.6:

Today's agenda is about disruptions. We'll talk about that a little bit. Mark Gordon in the previous presentation talked about Suez Canal disruptions and there's other ones that we need to consider, but we don't have to go too far back in time to think about disruptions. I'm going to put forth some concepts about AI coupled with scenario modelling, and I would argue that this combination is a really suitable and natural hedge against supply chain disruption. You'll know what I mean, because I'm going to bring out an example that you're all very familiar with. We don't have to go too far back in time to remember these. In 2019, we had the COVID pandemic, that severely extended lead times for both manufacturing as well as for logistics and also had a very profound impact on manufacturing costs. It's easy to think about how COVID affected us on a personal basis and not think about supply chain. So let me just talk about some of the facts during that time. Just in terms of lead times, ocean shipping lead times from Asia to Western Europe, say, Yantian, China to Rotterdam, went from 32 days, going through four ports along the way, to 70 days, and in China, more than 75 per cent of the plants shut down completely or were severely underutilized during this time. So naturally, this led to production halts, which led to extended lead times and also production halts and creating global shortages, in particular, of semiconductors. 

 

Jerry Holbus 0:01:40.8:

So the lead times for some semiconductors went from 13 to 26 weeks. Now in terms of production costs, or actually all costs, production and logistics costs, shipping a 40-foot container from Asia to the United States, which would normally cost about $2,000 went to $20,000 for a period of time, and because companies were thinking about different ways of managing inventory, from just in time to just in case, they tied up their capital and also placed a huge load on warehousing, driving up those warehousing costs. Now, two years later, we had the Suez Canal issue. There you see the Evergreen container line Ever Given, and even though that ship was stuck in the canal for six days and seven hours, it came at a cost, and that cost was $400 million per hour. So huge expense. Now, not all disruptions are negative. Some can be actually positive, like a big demand event that needs to be planned out through the supply chain. Here's an example of a construction company that had secured a major onshore gas pipeline, but by and large, when we think about other disruptions these days, we think about tariffs, and the news is fairly negative. There was an analysis done by the research group in Detroit for the Center for Automotive Research and in April of this past year came out with, wanted to understand the impact of a uniform 25 per cent tariff on all trading partners and what that would have on the automotive industry overall.

 

Jerry Holbus 0:03:27.7:

As you can see, or maybe can't see, they're expecting an increased cost of $108 billion for all US automakers. So this is some serious money and something that we try to hedge around as much as we possibly can. So we can conclude that disruptions in supply chain, they're pervasive, they're persistent, we don't know when they're coming, but they do come, so we need to really create the stance that we're prepared for them at any given time. The impact and the response and how we need to react can affect any planning horizon. So we think about a strategic planning horizon of three to five years, tactical planning horizon of six months out to 24 months out. The operational planning horizon, which is really time zero, going out 18 months. It's really dependent upon your longest vendor lead time, and then the executional time frame, which is zero to two weeks out. So let's take a look at a disruption that we deal with almost on a daily basis, whenever we get into a car, to just drive the point home of how you can go about mitigating the risk of a disruption. So this is pretty easy to understand. You've got a Google Map up here and we're, in this particular case, we're choosing a path from Stanford University to Berkeley. This example is to bring about how we go about predicting traffic conditions and also travel time, and I would argue that the method that is used by Google is in fact a natural hedge for all disruptions, but in particular supply chain. 

 

Jerry Holbus 0:05:13.8:

Now, you're probably very familiar with this application. I use it because I've got kids, and they need to go places many times - well, I would say many times a week, but actually it's many times a day - but probably what you didn't know is that this is a graph machine learning problem, and graph machine learning in a sense is used to create predictions. Now, the way the graph is created is, it uses nodes to actually represent the road segments themselves, and those road segments are tied together with the edges of the neural network graph. Now, as you can see under the Node A, there's many different colors represented there. That means you can have a large number of road segments considered for any one node, which makes sense, but it also suggests to you that this is combinatorically a very, very large problem to solve, and when we get the feedback from Google, we're usually just presented with three or four routes. So there's a pruning process that needs to take place. Now, what's also really profound about this is that Google employs something called an adaptive response. So that if it senses that any one of these small road segments is, and the traffic on that road segment is moving slower than normal, it will then plug in real-time data for that road segment or perhaps another road segment, and then it strings together those road segments to come up with the ATA, and it generates that in milliseconds and presents it by way of an API or it sends it directly to the app. 

 

Jerry Holbus 0:06:49.8:

So generally, here's how it works. The training data is collected from that anonymized data, but it could be training data from a year ago, and that by and large is used again, and again, and again for each one of those road segments that are created by Google. It's then, those predictions then are pruned by their master data, in a sense it's the Google production routing system, and what comes up then are scenarios that are then ranked by the ETA. So it's a combination of both graph neural network and AI with scenario modeling, and this happens, like I said, sub-segmented, it happens a lot, it's a very repetitive thing that's going on in the background. All right, so following through on this train of thought, disruptions are persistent, they're pervasive, the impact and response can be mapped to any one of these planning horizons. AI and scenario modeling are a natural hedge. We know this ourselves because we're using this technique almost every day, but disruptions may vary on the frequency and impact. So something like traffic time production or traffic time prediction or traffic patterns, something that we consider every day. So for those repeatable and those predictions that we want to create again, and again, and again, it's good to have an autonomous process for that, but others are a little more complex and they need complex scenario building and analysis and discussion with other key stakeholders. 

 

Jerry Holbus 0:08:28.1:

So let's consider this one, and I encourage you to check out the demo area because we have this particular presentation in the demos, and what this is doing is presenting how one would rationalize a tariff against a number of different countries. So you can enter the tariff rates for any number of countries there and compare scenarios, and once you pick a scenario, you then can extend the analogy by adding additional data, such as labor costs, against that one particular scenario combination that you wanna consider. Now this is all for one particular part. You can see up there in the upper right-hand corner, this is for the powertrain, but we can consider many different parts within an automobile, and an automobile typically is over 5000 parts. What would be interesting also is to bring it down at a lower level, maybe not just subcomponent, maybe we want to go to the commodity level, and maybe we want to analyze the country commodity combination like aluminum from Canada to see what the impact is. Now, to do this on an active basis, one might think not just traditional AI but maybe agentic AI that can go out do its own research on the combination of, or an understanding against an engineering bill of materials, how much aluminum there is for all the parts that make up a particular car. So it's a good experiment for agentic AI, and there are some companies right here in the Silicon Valley that are working on this today. 

 

Jerry Holbus 0:10:02.9:

Since I mentioned agentic AI, I might want to define this a little more practically so that you walk away with an understanding. So the traditional AI is very reactive. It's responding to a specific request and very much like the traffic time and road condition problem, it gives you a particular response based on the inputs that you provide to it. Now, agentic AI is a bit different. It kind of acts on its own. It does its own research. It makes its own decisions and takes an initiative and autonomously tries to converge to a solution and always look for a better solution once that happens. So if we see some examples from supply chain, demand forecasting would not be an issue that's solved with agentic AI. It's just not agentic because I'm providing some inputs and I'm expecting it to converge to a forecast. So it's basically making a prediction on my behalf. On the other hand, I could use agentic AI for exception handling, as an example. The traditional approach would be to present some alerts, and in supply chain, those alerts could be, hey, you've got a material shortage, or you've got some late materials coming in, or you've got a capacity overload, or a min on hand condition, and it would take a rules-based approach to solving these issues, such as, I want to prioritize allocation of inventory on the basis of how important is this customer, or what is the due date, or what is the size of the order, maybe I want to make sure I prioritize small orders over large orders to get more throughput, and so on. 

 

Jerry Holbus 0:11:45.2:

So with that, why don't we get into the Q&A and get our experts up here? So gentlemen, if you wouldn't mind taking a seat. Amar Sanghera, who is the worldwide leader for strategy definition and go-to-market for AWS supply chain solutions, and Abhijeet Shetty for the past ten years has been managing director and partner at the Boston Consulting Group. Thanks, guys, for being here. All right. So first question. It's all about AI trends. What are you seeing as the biggest AI trends in supply chain management and what are your comments on these trends? Are they sustainable? Are they durable? 

 

Amar Sanghera 0:12:35.1:

I think last time we had a conversation on AI trends, I think you mentioned a few of the words which are being used, but fundamentally if I look at AI, I think the realization and supply chain perspective, and it could be any part of the business operations, is that there is a need for augmenting human effort which is being taken and with the complexity being added by, I would say either competitive pressures, external factors, the need for making some of those insights to be generated, right, is the first part. Which means people are looking for inputs which they can synthesize much faster. So there's a huge amount of data which is already there. We saw in so many presentations earlier as well. There's a vast amount of data. You want to have silos of data being brought together, being linked together. So one thing which I would say is when we think of AI and trends, I would rather disambiguate and say, well, the broader trend is that the decision-making process and supply chain needs improvement, and that's where I think a lot of the investments which we see from either tech companies, which are investing in building capabilities like Anaplan and others, is going in, and also from a customer's perspective, as retailers, I represent AWS, which is Amazon internally, is one of our biggest customers, a big focus has been on eliminating what we call as a cognitive load on humans.

 

Amar Sanghera 0:14:28.1:

How do I make processes, I wouldn't call them autonomous, but smarter, that they are able to take and present simple decisions to be made so that the people can choose the right opportunity as it is being presented. So I would say for me, this has been something which has been persistent. The only thing is that now with the hype - let me call it this way - with the hype of AI, everybody is talking AI, gen AI, and others, it has just come to the fore. It has been around for a while, and it's just that the trend has now become mainstream. 

 

Jerry Holbus 0:14:47.4: 

It's been around since World War II, right? That's when the [?neural 0:14:50.1] program was first put into production. 

 

Abhijeet Shetty 0:14:55.6: 

So stepping back - Jerry, you already spoke about disruptions. Supply chain disruptions is not this year's phenomenon. It's fundamentally changed since 2020. That is going to become steady state or normal for at least the next five to ten years. We know that. So business leaders, not just supply chain officers, but COOs, CFOs, CEOs, are all fairly cognizant that supply chain investments need to go higher, and also at the same time, supply chain needs to become more dynamic. What that means is AI is only going to get more and more prevalent in supply chain, and that is something that we, as BCG, see in our work with clients as well. Incredible amount of interest in AI and all flavors of AI. There's machine learning, there is Gen AI, there's agentic, autonomous workflows. Incredible amount of interest. I think within that interest, then we need to parse out what works and what is at scale, and what is truly at the frontier. A couple of use cases that are already at scale and are working in the core of the business, network flow optimization, phenomenal use case for machine learning. Demand forecasting, phenomenal use case for machine learning, proven at scale, almost everybody has an offering that utilizes machine learning for demand forecasting. If I were to then go to a second bucket of what's at the leading edge and early at scale, IBP and end-to-end planning is becoming at scale for a number of organizations, using a combination of both machine learning as well as gen AI and some agentic techniques. 

 

Abhijeet Shetty 0:16:42.7:

Then at the frontier, what we are seeing is both agentic AI and early attempts at running proof of concepts or pilots to see semi-autonomous agents plus human systems, and then some experimentation on the absolute bleeding edge on embodied AI with human or robotics. So I want to leave that outside the remit of this conversation, but a number of clients are signing up letters of intent with start-ups who are doing work on embodied AI, many of the leading tech providers are in that space. That's very different. I think in terms of supply chain trends, there's stuff that's full at scale delivered value and then there's stuff that people are playing with right now. 

 

Jerry Holbus 0:17:30.5:

When people think about investment, they normally think about the financial investment that they have to put in, but it's also an educational investment and a time investment, too, that we can't discount or underestimate. 

 

Amar Sanghera 0:17:57.4:

I think one element I would just add, what Shetty just mentioned, I think element is around skilling of people. When we think of the trend as well. I think with the GenAI, with ChatGPT being pervasive and many other tools, like Google launched Gemini. If you go to Amazon actually, I don't know how many of you have tried it, there's Rufus which actually gives you recommendations. I'm sorry, Amazon plug. But all of that is now making people more used to trusting what you're getting. It is a double-edged sword in the way I look at it. When you think of AI, and I think Shetty mentioned it, and I think I'd love to get deeper into it. There is a predictive part of it, which is machine learning. There is AI, which is actual generating intelligence out of it, which is trying to mimic human in some way, and then there is the autonomous AI, AGI, where we really want to be. Agentic is a mishmash of what I would call as traditional RPA, BPM, plus AI, plus now newer capabilities which allow it to be almost like a goal seek because of the amount of compute power we have now. Thanks, Nvidia. I know there's somebody from Nvidia here as well. All of these things are essentially bringing that double-edged sword that we have to be able to understand how people are able to trust what the agentic AI or tools or AI is generating for you, which kind of like takes us back to one big element of trend. 

 

Amar Sanghera 0:19:10.1: 

While people are experimenting, not everything in AI and ML is foolproof. There are challenges in using AI-based decision-making. For example, if I say, hey, if my forecasting - and I'll give you an anecdotal example, large language model using forecasting has been played around for a while. There are LLM models which are trying to do predictions, for forecast. They have been actually, in many cases, inferior to the traditional machine learning models because they can generate random variations of hallucinations, but this works very well for certain scenarios. One thing I would say, bit of trend, a bit of a very latent, I would say in the minds of the technologists who are making this adoption is also giving out a warning that, don't go too fast. Make sure that something, whichever you do, is trusted, can really scale out before you start adopting those as an AI trend as well. 

 

Jerry Holbus 0:20:07.0:

That's a really good point. I was thinking about the traditional AI example that I talked about, which is the traffic time and ETA prediction. One way to make that agentic would be to rearrange, let's say, an executive has an AI agent and it's been determined that he's going to be late for a meeting. So the agent would then go out and either cancel the meeting or call or send a text message to the other participants in that meeting, saying I'm going to be late, let's reschedule for another day. That gets a little bit scary if it's not tightly controlled. Maybe he doesn't want to cancel the meeting. Maybe he wants other things to happen. So yes, there's a lot of imagined possibilities if you don't have those guardrails and that tight control set up on those agents. All right, let's go on to the next question. Most supply chain organizations have experimented with AI and have gone so far with agentic AI. So what allows an organization to scale and prove value? 

 

Abhijeet Shetty 0:21:10.9: 

So three or four things that are absolutely critical for scaling - oh, you've already got them. I think there was a fairly significant lead in your question, which is value. It is incredibly important to lead with value. There are no more checks being signed by CFOs for just play money to experiment or run proof of concepts. The market is now very, very aware, and industry leaders are very aware and educated on AI, and so therefore, they want to spend money where there is ROI, and the investment makes a return. So focusing, as an operator, on use cases that are proven to deliver value, and now there is enough experience in the market over the last four or five years on use cases that deliver value and use cases that are early to value. It's a fairly safe bet to invest on use cases that deliver value. Then having a straight-line track between what AI is going to deliver and where it gets landed on the P&L. So for example, if you are doing machine learning driven demand forecasting, the objective is not to improve forecast accuracy alone. The objective is to use that forecast accuracy to drive down inventory and drive up service levels. That is where your senior leaders and CSOs, and CFOs, and CEOs start taking notice and saying, okay, then it makes sense to ramp up and improve our investment. 

 

Abhijeet Shetty 0:22:47.0:

The second big thing is speed to impact. Gone are the days of two-year, three-year deployments. We are no longer in a business cycle where digitization can take three, five years, and then there is a long lead time after that of delivering value. You have to deliver value at a business unit or a category level, prove that value, and then start scaling up to the organization. So start small, deliver value, even if that value is not hundreds of millions of dollars, but a few million, but it's in a contained business unit, you can then start rolling it out. That improves value. It's not a proof of concept. You may call it a pilot, you may call it a late-stage MVP, but it's something that delivers value that you can then drive on to the future and scale. The last - Amar spoke about this - in BCG, we've got metric and a heuristic and a principle that says 70 per cent of the value of a transformation program, either investment, time spent, energy, is change management, and AI works best when it works along with humans. The problem is not AI, the problem is humans, and changing somebody's behavior and saying, you can no longer rewrite or overwrite a forecast, is extremely tough. You cannot go to a sales leader and say, you, I'm sorry, don't have a better perspective of how the market is going to change. The model has a better perspective. That's very tough and that takes time. 

 

Abhijeet Shetty 0:24:16.3:

That then again ties back to point number two, where if you are able to do it in a business unit, you can prove out value, convince the business leader. The business leader then kind of becomes your proponent, advocate, and champion in the rest of the organization. So focusing on that and then bringing an integrated team. AI is not just a technologist's hobby at this point in time. The business has to get core involved in it. We, as BCG, consistently advocate for a business leader, either as head of sales or the head of a P&L to take on the transformation program and lead it versus leave it to the IT organization or the supply chain organization, and then the team is a cross-functional team comprised of technologists, supply chain practitioners, even the finance person is in the room to monitor the investment and the ROI, and that's when this becomes a success. 

 

Jerry Holbus 0:25:17.7:

One of the things that - based on my implementation experience - that works in motivating people to take on a new technology is, sometimes you need to find volunteers within an organization who's really interested in taking this on and have them run with it. If they can be successful with it, and you kind of set them up for success, communicate the methodology that they use to other parts of the organization like you were talking about the rollout. 

 

Abhijeet Shetty 0:25:43.2:

Yes, absolutely right, Jerry. You have to find hungry leaders and that may not necessarily be the leader of the largest business unit or the most important business unit. Typically, we find leaders mid-stage in their career who are hungry and ambitious and have the energy to drive a program like this, in the smaller or mid-sized business units. That is where you look for the talent that will propel this into the organization and becomes a phenomenal way for the senior C-suite leaders to start spotting talent for the future. As we said earlier in the preamble, AI is going to stay. The people who take that challenge on the best and drive it forward are the ones who will succeed in the organization five years, ten years from now as senior leaders.

 

Jerry Holbus 0:26:32.6:

This next slide I wanted to bring to bear because it addresses a question that we were just talking about. How can you guarantee yourself success? Sometimes it doesn't make sense to implement a big bang project. It makes sense to scope it down, tackle a particular business problem that you would like to achieve. Maybe I want to decrease my operational expense and supply planning by 15 per cent by eliminating all expedited shipments. It's a natural business release. So the focus then is, you just need to find that functionality within that application or that AI system to work on that particular business problem and just say no to everything else. So you can build up and understand and master the methodology for doing that. It also means limiting the amount of data. It's functionality, it's people assigned a particular role, and a measurement system or a governance system that says yes, after three weeks of doing it this way, we achieve these kind of results. Then you go on to the next business release, the next business release, and before you know it, you've created a lot of value to the organization, and we used to do this repeatedly. This is more than 20 years ago. We would fund a company's ERP implementation because we took a different approach with the planning system that we were implementing. 

 

Amar Sanghera 0:27:53.6:

So, Jerry, just as an add-on, as a thought process, right? So one easy way which I think we can think of is working backwards from a problem statement. That generally is something very difficult to arrive at, especially given there's so many distractions, especially in supply chain area which comes, right? Oh, I want to improve my supply chain processes. There can be a motherhood statement. I want to reduce my inventory, but that's not really what we are looking at when we say working backward. Sometimes it is about decomposing your problem statement into something very, I would say, metric-oriented, which is actually impacting somebody's business process, actually driving an outcome and a goal, but then also making sure that you are able to iterate and experiment with the opportunities very, very quickly as a proof point, right, to be able to run it very quickly to generate the first set of, let's say, outcomes and actually see the impact of that value being adopted in a very granular way as well. A lot of times easier said, but I think one thing when I think of, let's say, the companies where I've seen AI, especially AWS customers where we've seen it being adopted successfully, have been the ones which have adopted this culture of bringing together IT team, a business leader, and an AI evangelist. Like somebody who is really into AI as like, I'm passionate about this, right?  

 

Amar Sanghera 0:29:23.8:

Not as AI first as a tool, that I'm going to bring this tool and try to do it, but as thinking and machine learning and AI and outcome-driven way, but then allowing that smaller team of five or six team members to really hone in on one problem statement which they can then quickly demonstrate to the leadership to say, look, I use this and I generated this. Now give me more impetus to scale, and then they are able to build a business case and drive those values. 

 

Jerry Holbus 0:29:50.6:

I like the idea of a cross-functional team. The word team goes back to a book I read by George Stalk in Boston Consulting Group who said that a team is people working full-time on a project. Anything else is a committee. So I thought, wow, that's really to the point. All right. Question three. From a people, process and technology point of view, are most supply chain organizations ready to take advantage of AI solutions in the market, what do you think? 

 

Abhijeet Shetty 0:30:21.6:

Maybe I'll go first. I think from a people readiness point of view, the readiness and willingness is there, but like I said, what we see in our experience is that there is significant upskilling to be done. For example, if a demand planner's job is no longer collating multiple inputs and forecasts and then deriving insight from that, because the AI models will do all of that. Then the demand planner becomes more of a business planner. Their job description fundamentally changes to say, okay, what are the potential risks built into the plan? What are the scenarios that I need to generate? How can I interrogate the AI model better to arrive at the right insights to take to the business leaders? Then how does that drive change in the trajectory of the business? So upskilling is a massive gap and will continue to be a gap going forward into the future. From a process point of view, AI programs are likely to fail when you're not reimagining the process entirely. We have seen when you apply AI to planning, for example, your planning cycle time can go from eight weeks right down to three to four weeks, because obviously AI speeds up the cycle, but if you're applying the same old eight-week process and using AI as lipstick on a pig to kind of make it look faster and make it look better, that's not going to work. We typically see that most organizations have a data swamp or a data lake, but they don't really have a genuinely working data fabric that feeds high-quality, consistent, high-fidelity data at low latency.

 

Abhijeet Shetty 0:32:05.5:

That is also a challenge because if AI is the engine, then data is the fuel that fuels that engine, and so we typically find data as a massive failure point or a critical failure point at the start of such transformations where you have not integrated the data well. Somebody in the organization says, we have got a data leak. We go in, start running water through the pipes, the data gets fed into the AI models and it starts spitting out garbage. Then people say the model is wrong. We go back and look at the data and realize, of course, the data was wrong. Shipment data is in four databases. It's exactly the same data, different snapshots, different granularity. Somebody agrees with it, somebody disagrees with it, and so you have to sit down. That data then becomes suddenly a human problem. You have to sit down with all the leaders and the operators and say, okay, which is the right source of truth? That is when you truly realize that data is a major issue in the organization. So I think that then completes the point of view. Amar, I'm sure you've got a number of [over speaking 0:33:15.6]. 

 

Amar Sanghera 0:33:16.3:

Again, I've maybe got a slightly different twist. I would say when we think of AI solutions, it's a very broad term again. It can be, hey, I want to do forecasting. Okay, yes, there's going to be a change and impact to one team member or a set of profile of people who are doing certain things and now they can do things faster, smarter, more efficiently. I know that word was used earlier in the presentation as well, efficiency, that transformation. When we think of the possibilities which we are looking at, I think this is a fundamental way that rather than thinking process first, this is our opportunity to really look at the outcomes and the impact you're delivering. Again, I'm going to use the same AWS and Amazon terminology we use. We essentially work backward from what is impacting the customer, what is the outcome I'm looking at? This is actually a bigger opportunity when I think of AI, because earlier, AI was a lot of times, the traditional AI or machine learning was in pockets, very specialized, is going to do something really, really well. With agentic and the ability to stitch those together - now again, making that assumption, if you have the data, what are the things you could achieve, assuming you have all the right foundations in place. Rather than thinking that I need to do this process because this is how I have always done it, this is a bigger opportunity where the business leaders, and that's why I feel many of these transformations need to be driven by business leaders and the business owners to say, I want to reimagine my, let's say, merchandise financial planning process. Why do I have to do it like this? Why do I have to operate like this? 

 

Amar Sanghera 0:34:52.9:

This is a bigger opportunity right now to really reimagine and break the mold and say, I want to arrive at that outcome and then I want to be able to assemble my business processes after I define that, after I understand how can AI and ML and different components drive that part. Just as an example, again, from my Amazon perspective, when we design many of our internal solutions and internal way of our operating, we start with that outcome or impact to - let's say, raising purchase orders. Vendors are going to get my purchase orders. I work backward from that and then define how I'm going to operate so that when I think of a solution, it doesn't become that I have to have a specific process in place first. I go with outcome and then start defining where all different automations, what are the absolutely must-have human inputs I need to build, and then build all my capabilities around that to say now I will be able to define a process. Then the persona of the people you want to hire, the type of trainings you want to give, the type of metrics and goals you want to build for the organization can be defined by the outcome rather than by a process first and the team first.

 

Jerry Holbus 0:36:07.5:

Greatest opportunity for AI-enabled supply chain planning? Did we hit this one or do you want to reiterate some points in here? 

 

Abhijeet Shetty 0:36:13.4:

I think the only thing that I would say is given everything we are seeing from a disruption point of view, planning will end up being a competitive advantage for companies and organizations. If you're not investing in planning and in having faster, better decisions informed by AI, at some point in time, you'll be left behind because everybody else is doing that, and especially in an environment, both macro and micro, because there is incredible demand volatility. There are supply chain shocks, and then there is geopolitical instability at the same time. So it's a perfect trifecta of multiple factors that are causing a perfect storm, which is not going to go away. This is a five-year, ten-year storm again, and so therefore, planning can easily and will end up being a source of competitive advantage going forward. So those who are not investing in this right now, I wish them good luck because there's no other way out. I think that's the biggest opportunity. So I wouldn't say the greatest opportunity for AI-enabled supply chain planning. I think supply chain planning is the greatest opportunity within AI in operations. 

 

Jerry Holbus 0:37:30.3:

Thanks, everybody. Thanks, gentlemen.