Episode 24: Assessing the Supply Chain Business Value of LLM's and ChatGPT - What Should Businesses Anticipate in 2024 and Beyond

Speaker A: Hello everyone and welcome to another episode of the Supply Chain Matters Podcast hosted by Bob Ferrari. Supply Chain Matters has consistently been recognized as one of the top Internet blogs in the field of supply chain management. Our goal in this podcast series is to provide the best insights and thought leadership from guests experienced and knowledgeable in supply chain management. Areas to be discussed include business processes, advanced technology and various industry and individual business, product, demand and supply networks. And with that, here's your podcast host, Bob Ferrari.

Speaker B: Hello to all listening and welcome to episode 24 of our supply Chain Matters Podcasting series. I am Bob Ferrari, the founder and Managing editor of our blogging platform and I serve as moderator of this podcasting series. The Supply Chain Matters blog is consistently rated as one of the top blogs in supply chain management thought leadership and since our founding in 2008 we have garnered a global wide readership. This podcast medium serves as a supplement to the various published content on Supply Chain Matters and is our opportunity to present a more two way conversational discussion to important topics and surely this one will be in the coming year. Businesses and multi industry supply chain management teams are now turning their attention to their plans for the year 2024 and in conjunction with our research arms 2024 predictions for industry and Global Supply Chains advisory that we will publish in January. In 2024 we are featuring a series of thought leadership guests that will be sharing perspectives and insights and on um, what to expect in specific supply chain management, business process, technology areas in the coming year. So I am very pleased to have as our guest for this episode Fred Lalayo. Fred is an entrepreneur at heart and Silicon Valley veteran. He provides an impressive track record building successful startups and driving technology innovation. Fred is a thought leader on the future of work and decision intelligence for the enterprise. He is a technology and startup advisor to hedge funds and financial institutions as well as an investor and an active board member for several startups in the US and Europe. Fred serves as the CEO of AERA Technology, a company described as transforming the future of work and business agility, leveraging decision intelligence. Prior to launching era he served as the CEO of Anaplan, which he grew from 20 to 650 employees and a $1 billion plus valuation. He also has held several executive positions at SAP Business Objects and ALG Software. Now I have known Fred for several years and I always look forward to our discussions on AI technology deployment in supply chain process areas. I am very pleased that he found the time to speak to our Supply Chain Matters audience and share some really good perspectives on this noteworthy topic. Welcome Fred. It's a pleasure to have you here.

Speaker C: Bob, thanks for having me back. Great to be with you today.

Speaker B: Good, good. So let's get into this because this is a really interesting topic to discuss and I'm sure we could fill uh, a lot of time about it. So from our conversations over the years, I know that you launched Aera Technologies decision intelligence platform before the AI hype that is currently dominating business and tech media right now. So let's begin our conversation with initially helping our listeners to better understand in layman terms, the fundamental differences between application of artificial intelligence and machine learning technology. That has already occurred in many areas of supply chain related processes with that of uh, what is now described as large language models or the familiar term chatgpt.

Speaker C: Yeah, so maybe we can break it down in uh, the different uh, aspects of AI. But if you think about the higher level definition of AI, it's the ability of computers to mimic human uh, brains. Right. So when we talk about how it applies in general to the world of supply chain, we've got to put that in the context. And the context of supply chain today is that digitization, uh, of our economy is driving uh, an explosion in the volume and the complexity of the decisions that uh, supply chains have to make to remain competitive. And that leads to leveraging, uh, AI, leveraging machines to actually do a lot of the work, make and execute a lot of the decisions that were historically, uh, made by humans, supported by machines. And it's also the ability to leverage this technology to make decisions that didn't have to be made, um, uh, just a few years back. So I call those decisions that were born in digital. So when we break AI again, as I said, the ability to uh, replicate what the human brain can do, you've got multiple categories, multiple techniques. Um, if you think about machine learning, it's an application of AI that uses data and algorithms to enable machines to learn and continually learn, uh, and improve based on experience. And they can do that without the human support. So they look at large data models, they train, uh, the right models, and then the AI is enabled, uh, to actually digitize decisions. Right? So you're looking at structured data, you deploy these algorithms that can help project, that can help predict, uh, and then you have all the types of algorithm that helps you optimize and you think about the decisions that you do. It's a lot about predictions, a lot about optimization. Now you have uh, a new type of technology called Generative AI, which is used by ChatGPT by MidJourney and by Other similar tools that are built on top of large language model as you mentioned and LLM have been trained on vast amount of data uh to understand and generate content. So one technology helps you predict based on a normalized content, the other one helps you generate content uh, based on uh, a large language model. So uh, the generative AI models at ChatGPT, they're designed to generate new data, new content and response based on patterns they learn from initial training data. The challenge is that in supply chain you cannot improvise, you cannot hallucinate. I would like to say your supply chain cannot hallucinate so they can generate text for various application uh, but the lack of structured data processing capabilities uh, that are very specific to a business, to a market knowledge, to a domain expertise, uh, makes the ability to deliver uh, reliable recommendations by those models uh very difficult or unreliable. You use machine uh learning, you use statistical forecasting, you use optimization to reliably help predict and optimize your uh supply chain resolve complex problem. You use large language model to build an interface between those algorithms and the users. You can now use that. You mentioned ChatGPT, you can now use that new technology to build an interface between a human operator and those underlying models. That interface is very open. It changes everything because instead of having to learn how to use a complex piece of technology, the large language modeling uh, trained on your environment can allow you to build that very fluid uh, interactions between you and the software. It's going to change the way you work, it's going to give you immediate access to content that was considered potentially expert content. But I would not recommend using uh a gen AI to run your supply chain, use structured data, use machine uh learning, use ah statistical forecasting or an optimization to come up with the numbers and use genai on top to interact with the users

Speaker B: to follow that along. Uh Fred, um, the word chatbots is commonly used when describing ChatGPT and what it is um, help our listeners a little bit about what is the meaning of chatbots when we talk about ChatGPT.

Speaker C: Well bots have existed for a long time Bob. When we launched Aera uh six and a half years ago, ChatGPT did not exist. We created Aera as a digital assistant that will understand how your business works, make real time recommendation, predict outcomes, uh, predictions coming up, optimize but also take action and execute the decisions back into your transactional system. So a lot of value in coming up with better, smarter recommendations, but also in decisions but also the ability to execute. And we created the era with that interface which is a chatbot type of interface where I can use language to ask questions and the system will translate that question into to be very simple code that will then retrieve the right answer from your normalized data, um, model. That's the way chatbot works. What um, ChatGPT brings to the mix is the ability to have a much more fluid interaction with the system. So the ability to create that code that is going to retrieve the data from your calculated data sources, uh, is now a lot more fluid. In the past I would have to ask very precise questions. How many recommendations do I have for that capability in that region for this week? If I got it wrong, the system would say if I got the orders or the syntax wrong, the system would probably say I don't understand. With the large language model and uh, those new uh, gen AI capability I can have a lot smarter interaction. The other thing that the ChatGPT would do and the smart bots will do, Bob, is you ask a question, the system will automatically know what information to present to you. Now again, the information is not created. I don't recommend that you leverage Genai to create the data. The data has to be calculated in your core systems in your decision intelligence layer. But it's very interesting to see a system that says you wanted uh, to understand what happened in my uh, um, uh, transportation lanes in Germany last week. I'm not only going to give you the answer, but I'm also going to show you the exact visual that is going to help you understand the answer. It's going to start learning as well from your interaction with the tool whether this was actually an optimal representation of the data, uh, for you uh, to make a decision. There's a lot of value. Don't underestimate the value of that. That new uh, layer of interaction is going to profoundly change the way you interact with the system. Um, but I would say the analogy I can use goes back to the GPS and the map. Right. Uh, the GPS is not leveraging Genai to create the localization of the cities and the towns and the roads on the map. That is hard work that's been done before. But you can now use an interface on top to say get me there. But the there is, you know, the coordinate of the there have been calculated uh, by another system ahead of time.

Speaker B: Yeah. And while we're on that topic and the learning and so forth, um, from my readings about this and my discussions with others about this as well, right now, uh, the promise of LLMs and ChatGPT has really caught the attention of Business leaders because you know, they look at it and they see the potential for profound changes in the way companies make decisions or the way work can get done eventually and whatever. But on the other hand, if you look today at the notion of what are the concerns, what are the shortcomings of the technology, uh, the notions of hallucinations that you brought up and so forth, there's a lot of things that have to be ironed out, so to speak, you know, to get this, all of this sorted out so we can get a little bit more into the mainstream aspect of that. Just from your gut feel, what do you think is going to be that timeline right now?

Speaker C: Ah, it's, it's a good feeling. It's also the conversations that we're having several times a day with supply chain leaders around the world. Um, it's now I think, I think the, the, you're absolutely right. I think what happened in 2023 is the chat GPT, the ability for anyone to go online, ask the questions and get beautiful text shaped up just shook everybody's uh, perception of what, of what is possible. Um, so I think you're seeing today executives, every board, every executive that we talk to is trying to get their uh, uh, gen AI strategy in place. Now as I explained with your first question, Gen AI will profoundly change the way you interact with, with your technology stack. Right? It's going to fluidify everything from retrieving information to executing faster. But the calculation, uh, the optimization that is required to run an efficient supply chain will be leveraging other, either AI or non AI technology. So it's a mix of things. You leverage machine learning, you leverage heuristics modeling, you leverage basically large data and analytics capabilities to come up with the optimal answer to any disruption in your supply chain, whatever it is. And then you leverage uh, uh, the gen AI to actually interact with the tool. The technology is ready to do the latter. What I've just talked about, that interactions with the operator, we have built it, uh, it's ready and it's quite powerful and I think it's going to democratize basically access to intelligence. We talked about 20 years ago before digitization, access to content, access to data. That's not even a question. Today everybody has access to market data to everything. Now it's going to be Genai would just democratize access to that digital intelligence that will allow you to demultiply the ability to make timely and accurate decisions uh, for your supply chain. So the impact is massive. I think the timeline in terms of technology readiness is There And I think the timeline for uh, massive interest and curiosity and willingness to invest which is really important is also there. But like always there is a concern around people just throwing under Genai everything. So breaking it down. Artificial intelligence, the ability of computers to mimic human brain. Within uh, this AI you'll find Genai which generates content, you'll find machine learning, you'll find decision intelligence which is what era ah specializes on, which is the ability to digitize and execute decisions at scale. You'll find composite AI, you'll find multiple AI. And if you think about it um, the areas that impact supply chain are Genai and decision Intelligence and gartner in their July 2023 Hype Cycle position both technologies as uh, transformational and coming to maturity within two and five years from now. Which means that large companies have already started when they talk about maturity they talk about everybody's doing it and the timeline is incredibly short. If you think about two years it's 700 and something days. It's nothing actually. Yes. Uh, so this is the train has already left the station and I believe you'll see over the next few weeks and months the talk back, the hype cycle, the marketing hype is going to start to fade away and real use case, real applications, real technology will start to uh, not emerge but uh, demonstrate massive values. Uh, and all this thing, all this conversation Bob is all about removing human based inefficiencies from supply chain. It's about breaking the silos finally. Not the data silos but the decision silos. It's about coordinating decision making at scale. And we're seeing with our clients today the cash cost service level but also carbon water. The impact of uh, uh the gains that you have when you deploy this kind of technology and you remove the challenges of human biases, human led execution which means potential mistakes. Uh and you also leverage digital to make faster decision closer to the point of impact. Many more decisions faster, closer to the point of impact. What you get, you get massive gains in your supply chain. This is the next unlock and I think that unlock will be absolutely huge and we're seeing it with our customers today.

Speaker A: Good.

Speaker B: And that's a good segue into uh, what have been the learnings that uh, companies have acquired thus far just using decision intelligence.

Speaker A: Um,

Speaker B: I know supply chain teams right now have been leveraging aero decision cloud to automate those areas that you mentioned about decision making across operations and planning and whatever. And you recently had the what, what was described as the ERA Hub 23 conference which I was able to view on live stream where customers really did a great job of describing their approaches and their journeys. Journeys and their key learnings, uh, especially related to uh, data management and what it really was, what they thought it was and what it ended up to be and how decisions were made and so forth. Um, and what were the unanticipated challenges? Could you share briefly with our listeners what were the most important learnings that you heard there and that you've heard even in your travels with customers and prospects about this?

Speaker C: Yeah, look, first of all thanks for joining the conference. Glad that you were able to join us. But I would say um, there are some learnings from our clients. Uh, the first one is uh, get started. And I think uh, we heard that from many, many clients was like it's scary, it's a bit daunting, it's moving to the unknown, but you have to get going. Uh, I think I love this uh, this quote from Unilever from Wendy Herrick who talked about moving from uh, uh, thinking digital, doing digital to being digital. And that was incredibly powerful statement. And the way they've reorganized their company, they completely structured a new customer operation, uh, office. They've organized themselves in order to be digital. And being digital means allowing digital to make and execute decisions on their own. Today we passed 10 million decisions executed last quarter. You talked about data. We brought in uh, more than a trillion rows of data, fresh new rows of data inside the platform. So you know, between the time you and I started talking about a few years and today it's becoming an operational reality. We see ERA being a, a core engine of supply chain execution. And uh, so being digital thinking with this is not just a decision support system, but it's a decision automation capability. I think Dan Vesset from IDC mentioned that. He said a few years back we were talking about decision support, then the word decision disappeared. We only talked about data, data Lake, data warehouse, data science. And today the word decision is coming back. But not as decision support, as decision automation, as decision intelligence. Um, I think that was really, really, really interesting. And two other thoughts. The first one is accepting that there is no alternative. From the point I made earlier about if you want to get the next step change in your supply chain efficiency, you're going to have to rely on this computer system, on this AI based system. Um, and for that you have to build a culture that is okay to fail. And I think that came back from Univer as well. Uh, creating that culture is super important because you will learn, you will fail. When I Look at the clients that we work with for so many years. Uh, some things that we started a few years back are not relevant today. Uh, and some things that we couldn't think about today are uh, sorry, uh, then are now hyper important today. So that flexibility in your mindset, that flexibility in your organization, the ability to fail. And as I said, we hear that a lot. It's always a little scary to start, but once you've started and you start building that approach in your DNA, uh, you don't um, regret it. So I think uh, it's really important to get started moving, uh, to start moving quickly.

Speaker B: Yeah. And you know what caught my, really m, caught my interest was the IDC data that indicated that 50% of data loses value within an hour. Yes, I've never seen that quantified yet, but that didn't surprise me. And then the other stat that they shared was 55% of top level managers don't understand how lower decisions are actually being made in the organization. And you know I looked, I looked at both of those stats and I said gee, we still got a lot of work to do in the data sphere, right.

Speaker C: So, so earlier in the conversation I mentioned that companies are facing a situation where there are new types of decisions that they have to make today that they didn't have to make a few years ago. If you're a company that advertises a lot, you used to you, you know, print or plan your campaigns, buy the space, print your material and here you go, you have a billboard on the freeway. Today more than 50% of that media spent has shifted to digital. So now you can activate or deactivate a campaign on one click. And by the way, you can measure the efficiency of that campaign on what's on the pixel level, which is like a square mile by the way. You can now correlate that to your own self availability. So you can say I can stop a campaign because I'm not going to have inventory or I can accelerate or increase a campaign that's performing because I've got excess inventory. Those kind of decisions are only relevant if they are made in real time and if you make them the next day, you've lost the opportunity. These are new types of just one example, but it's an easy one to understand. These are new types of decisions that were not needed before. So back to your point about IDC's quote, 50% of data decays within 24 hours. But now I think they're, don't quote me on that number. It's either 75 or 65% of decisions have to be made within a day, otherwise they're obsolete. You can't make them anymore. So that when I talk about the acceleration of business decision making and increasing the volume and complexity, we're right there. And I think IDC did a brilliant job at framing the conversation there.

Speaker B: Yeah, yeah. And then one other thing I mentioned was uh, Sascha Koff, um, SVP at uh, Dell. And what striked me about how she described this was the notion that they looked at all the decisions that were being made, just as you said, um, and what did they apply to, what goes into the decision, who's involved in it, who needs to be involved with it, how important is the real time aspects of the data. She raised all of that and I thought that was really insightful.

Speaker C: Dell has done a brilliant job in deploying decision intelligence um, quickly and efficiently. Um, and the very interesting talk about the evolution of the topic of decision intelligence, Bob, years ago we were looking at uh, what is the one use case we're going to apply it to? Is it going to be a demand forecasting, is it an inventory rebalancing, is it a, whatever, the oldest order allocation, whatever use case? Today we're seeing most of our clients and new clients taking a platform approach. I'm not talking technology here, I'm talking about uh, they're looking at, just like Del mentioned, they're looking at what they call frictionless supply chain. So across the supply chain, across their supply chain, they want to digitize all the decisions that can be digitized. So it's not one, it's not two. Same thing with Unilever, same thing with most clients. You can see now when you look at their architecture, there is a box that sits on top of your ERP, that sits on top of your systems of uh, differentiation, your planning tool, your wms, your M. Tms, all these tools. You now have a layer of decision intelligence that here to automate and augment your decision making, working with your operators and that approach. I think Dell is one of the leading example of going across and not focusing specifically on one use case. I think they understood things, but they've started on that journey. As Sasha mentioned three years ago, they started on that journey and we had the opportunity to start partnering with them about a year ago. So they build that experience uh, in how to do it successfully.

Speaker B: Yeah, and I think the important takeaway there is that even for a company like Dell, it was a three year

Speaker C: effort to, oh, and it's not over, it's A transformation of uh, you're transforming your people, you're transforming your process where you bring the two together. It's really changing the way you work. And Dell is one of the first that uh, went pretty much straight to automation. So you think about decision intelligence. You have a system that delivers recommendations to say, Bob, I recommend that you uh, repoint this order from this place to that place within the next 12 hours in order to avoid blah blah, blah. Well that system can do it by asking you for your decision based on that recommendation or the system can analyze the outcome of that recommendation and execute it on its own. Full human out of the loop. Um, and I think Dell was very quick to say we're going to go faster with the human out of the loop to enable humans to actually uh, work on other topics. So very interesting, uh, deployment case here.

Speaker B: Absolutely, yes. So before we run out of time, I do want to touch about one final topic. Um, and that is can you provide our uh, listeners an example how Era itself, uh, the company is approaching LLM and ChatGPT to enhanced uh, supply chain decision making in your platform. And are those approaches, are they customer driven or are they product driven?

Speaker C: Uh, it's a double, double. Very good question. I will start with the second part. It was uh, product driven. We drove that um, we are a little bit ahead of the market which is where you want to be as a technology vendor. You don't want to be left behind. So as I mentioned when we at the conference, Bob, that you were uh, kind enough to attend, you saw me present that video with the self driving enterprise that's actually six years old and it looks like sci fi then but it looks like the reality today. And uh, we've been on that journey for quite some time and we knew that our technology was going to be ahead of the market. But now the market post Covid, I think it took about a year for organizations to kind of get their banks back and, and be able to kind of rethink the next generation of technology that will change the way work gets done and we're at that point plus the technology is ready. I mean you use ChatGPT and it's amazing, it works beautifully. So our ability to combine that technology with the core engines that we built to digitize and automate decision making, um, I would say it's been pushed by era, but I think that we're going to be shifting to a pool market. Uh, from our standpoint like now it's starting to happen. The questions are coming. When we presented ERA speak to the phone like I showed you several years back, like Syria Alexa and ask supply chain question to get an answer. People were looking at us going like this is cool but this seemed a little bit advanced for us. This is not the case anymore. I think the market is moving very, very quickly from what I call a push to a pull. And to answer your first question, we have ah, built. We see that coming. We're in Silicon Valley where we are trying to be on um, top of everything. And the one thing I tell you is that if your technology architecture has not been designed with this uh, in a future proof manner, you're going to be in serious trouble. Because the speed at which the um, innovation is happening in the field of generative AI is mind boggling. We're not struggling to keep up because we've got a great team and we're on top of things. But it is absolutely amazing. And everything I've said around today, uh, your genai be careful because you don't want your supply chain to hallucinate given the speed at which things are moving. Um, you know it's possible that a year from now there will be new improvements and new guardrails and new aspects of that technology um, that will resolve the problem. One of them is autogn that just came out that allows the synchronization of multiple gen AI models. I mean I can't go there right now but there's so much going on. Um, I would encourage everyone to really uh, stay connected. Um, uh, and it's the first time in my career, you mentioned my career earlier. I've done this for 26, 27 years. It's the first time that I feel like if I let a week go by without talking to our engineers and without reading and staying informed, I might be disconnected. So the speed at which this innovation is happening um, is mind boggling and uh, it's very exciting. Um, but you have to have a future proof technology and make sure that the partners that you're working with are staying uh, on top of everything that's going on the net Net. As my conclusion is, all of that is great but it's not technology for technology's sake. When we look at the impact of this technology on um, the core efficiencies that you can gain on supply chain and there's thousands of trucks that can be taken off the roads, the millions of miles that are not driven, the tens of thousands of carbon, uh, tons of carbons that are not emitted, the raw material that doesn't get destroyed. Those examples are coming through my mind because I'm talking about customer examples here. That's the exciting part because I think it's going to be, as I said before, the next step change in sustainability and sustainable supply chain, which has been uh, a goal but very hard to achieve, uh, uh, for top of mind for a lot of people for a very long time. And I think that technology will help, uh, change things once and for all.

Speaker B: So as always, Fred, a very, very insightful conversation. I think you've helped our listeners, ah, better understand what's going on with uh, LLMs and ChatGPT or LLMs versus ChatGPT, depending on whatever context it is. But thank you again for taking the time to speak to us. I really, really appreciate it.

Speaker C: Um, it's a pleasure Bob. Thank you for having me.

Speaker B: Yes, and if any of our listeners want to contact you, how can they get a hold of you?

Speaker C: Era technology a. Era technology.com and then you mentioned our conference. I think it's still online and available on demand. It's era a era hub hub23.com so, um, um, you can hear from the folks that we've just uh, uh, mentioned in the, in the podcast today.

Speaker B: Very good. Okay, thank you. So this concludes our Supply Chain matters part, episode 24, assessing the business value of LLMs and ChatGPT. What should businesses and supply chains anticipate in 2024 and beyond? Stay tuned to the Supply Chain Matters blog for additional announcements as to upcoming guests and compelling topics related to what you heard today. Supply chain business processes and decision making needs. As mentioned in our introduction, we are also lining up other thought leader guests with their perspectives of what to anticipate in 2024. And stay tuned for that as we queue up a little bit more of the speakers there. Feel free to contact me with your request for additional timely topics. And in the meantime, this is Bob Ferrari signing off until our next episode. And always remember, supply chains do matter for successful business outcomes.

Speaker A: Thanks for listening to this episode of the Supply Chain Matters podcast hosted by Bob Ferrari. For further information and insights, please Visit our websites www.theferrarigroup.com or the Supply Chain Matters blog at www.suppply-chain matters.com. thanks again for listening and goodbye.

Speaker B: It.