Jeff: Hello again everyone. You're joining us for another episode of executive Platforms Blueprint podcast series. My name is Jeff Nicks. I'm head of content and research. My guest today is Chris Cutshaw. He's with TMC, a division of C.H. Robinson. We're going to be having a conversation about generative AI, the role that it's going to play in orchestrating supply chains in the future. And he's recently given a presentation on this that actually gets into some specifics, which I'm very excited about it. I think most companies are very interested in this space, but a lot of it sounds theoretical. So if we can actually get into weeds. Yeah, I'm very excited for this conversation. Chris. Thanks so much for joining me. Chris: Yeah, no problem. Thanks for having us. Jeff: Why don't we start off just to make sure we're using the correct terms here. When you say generative AI, what differentiates that from the broader AI category? Chris: Yeah, I think a few things, right. So people have been saying AI for a long time, and I think with some of the innovations that have just happened, specifically open AI, where this content is actually being generated. So it's in a human readable fashion. It's able to create narratives, create language that we can understand and interact with. So that's where the generative comes in. And really, since November 2022, when open AI launched the chat GBT platform, there's definitely other platforms that are coming on board. It's really made it tangible for usage and supply chains. I think is ripe for usage in the space. Jeff: You gave a presentation a little earlier today that was actually geared towards an audience of senior supply chain executive saying these are some of the possibilities. This is what some people are already doing with it. I'm not going to ask you to repeat a 35 minute presentation, but what were some of the key takeaways you hope people got from? Chris: Yeah, key takeaways are it's definitely very new. So we're in the learning phase, but I think the opportunities are just tremendous in our space and supply chains in general. There's a lot of partners. There's a lot of people involved, a lot of different systems. Those things don't always harmonize together, and you end up having a lot of people doing manual things, processing pressing buttons for, for lack of better terms. So if you can take that value added work or the non value added work out of their efforts and actually focus them more on things that can be continuous improvement for their supply chain, things that can optimize, that's where you want to start focusing some of this new technology to take that tactical day-to-day work out of their hands. So they can focus on improvement ideas. Jeff: I can also see this being a powerful tool for giving data a voice because I think a lot of what digital transformation talks about is there's this, you know, cascade of data coming in and you have to figure out what's actionable if a an AI tool can actually put it into words. This is why we're doing it this way. Chris: Yeah. What a powerful tool. Jeff: Definitely very powerful. Chris: The opportunities are just starting, but they really seem truly transformative and to your point, you know, when you have a lot of data, you have a lot of information coming in. It's tough to process all of that and make intelligence out of it, and I think technologies like this are starting to actually interact with a definitely a diverse set of information. Taking information out of that and making it tangible so that you can automate processes I think is really exciting. Jeff: Yeah. So again, I try to sometimes put myself in the position of a supply chain leader who is aware this tool is coming online. I certainly don't want to be the last person to look into this, but it can be intimidating to be first. What are some of the low hanging fruits out there that maybe I could experiment with or that I can watch someone else you know be an early adapter and learn from them? Chris: Yeah. I think first and foremost, you need to make sure you have your data security and data privacy completely solved, so you want to make sure the way you're using this doesn't expose information that it shouldn't, and that it's really targeted and focused. So some of the low hanging fruit that supply chain leaders know is, you know, we have people doing things manually across the value chain or supply chain. So really saying hey, where am I spending the most amount of time? That's not value added work or increasing the value of what we do for our operations and kind of aggregating and stack ranking that and seeing hey, do I have the data that is enough to solve this use case and then can I stream that to a confined AGI agent that potentially can take over some of those actions. So starting very small to say, hey, I'm pressing this button or I'm responding to an email question that is really manual nature or tactical nature. And I do this a lot or I have teams of people doing this. Can I start there to remove some of those people or more importantly, transform what they're doing and uplift their talent to more, you know, innovative and value add items. Jeff: I was interested at the beginning of your answer. You saying you really have to lock down? What data you're giving it that it's the right data that it's secure. I can see that being something that, you know, companies want to be very clear. What are maybe the new rules involved? I mean data security is always something we talk about, but we're actually feeding and I don't want to be dramatic information into a black box where it will do something and then an answer is going to come out the far side. What does getting that data ready look like? Chris: Yeah. I think really when you're doing working with AGI, it's all about prompt engineering, so you you're training the model or the agent to respond. Specifically how you want it. So really taking in time up front to make sure prompt engineering is reducing the ability for the API to access information that it shouldn't, or the ability for it to create narratives that let's say are making things up are fabricating information. So you really wanted define it, say. Here's your role in life. Here's what you're supposed to do, and by the way, never fabricate information. Follow a set rigid process, but what's really cool is what we've done is actually putting planning large language models in front of that, so that you really create a specific agents that have tactical, you know requirements or tactical options that they can go after accessing internal tools, really confining the data to make sure that it is really exactly what they need and know more and then making sure the outputs are in a very rigid fashion. So it's it's controlled on the use case that you're focusing on. So again, stepping back, looking at all of the manual processes that happen that you know are redundant and that your teams don't love to do taking that and starting your journey within open AI type of technology or there's some other ones coming online and focusing really tangibly at some of those use cases and going after them. Jeff: I'm so interested you mentioned sort of teaching it that it can't fabricate. So I and forgive me, I'm a little more familiar with chat, GPT than some of the industry applications, but I know hallucination is one of the terms that we've heard. Where it just wants to please you. It's not necessarily giving you the right answer. It's trying to generate a response that you're going to be happy with. Me like that was the right answer. Whether there's depth to or not, this can't be that because it's actionable data that it's chewing on. Can you expand upon that one? Chris: No hallucination is actually a technical term for some of these AI agents. I mean, there's examples of law firms using this and not checking the output and found out that, yeah, citations making up cases, right? So you have to be very rigid and clear and the prompt engineering that you know you need to follow and make sure that you're only speaking on truth, their data that you have accessible to and when we're interacting with like some of those prompts that are open, they can go a lot of different ways and what we're focused on is deploying that to our code base using some of that trained model to respond in human like way but really focusing what it can do. So that doesn't go outside of its boundaries. Jeff: I'm so interested in this and I remember you saying before we started recording that you do actually have a few details and hands on things that we can get into. I'd love to do that now as much as this big picture is fascinating. What are some concrete examples of someone has applied this to a? What is it doing for them? And you don't have to name a clients name, obviously. But like walk us through the scenario. Chris: Yeah. Well, I can give you one scenario very specifically that we started on it and have in production. So a lot of times in our industry, which we're really focused on, transportation is when you physically deliver something, someone signs a document and it's called a proof of delivery. And for a lot of customers to automate their payment process to ensure that it was on time and in full, OTIF really focused. You want to capture all those documents, and if there's an audit that happens, you want to make sure they're right inaccurate. So across, you know, thousands of shipments that you're managing, you're dealing with tons of different providers and those providers have sub providers like drivers or warehouse workers. So you have all these sources of information where documents can live, and often there's teams of people that go chase and send emails and follow up and get that data back on system. That's a perfect, tangible use case that you could focus an AI agent on instead of having, you know, teams of people go after that. So chasing PO's chasing documents, following up, you know, when something's not right and getting the information in an automated way that you know you've set up the boundaries and control that are taking, you know that manual labor out of the equation and focusing their time on more, let's say, optimization activities that can reduce cost or improve service and they're not getting bogged down with some of the tactical day-to-day work. So I think you know documentation and sharing the documentation is accurate, timely and it's associated to the shipment record is where we've we focus some of this build out on. Jeff: Fantastic. So to maybe say that in other way and you can check me it can produce the report out of all these data points. But what it's freeing people up there used to be someone who had to compile all that, who had to make it make sense in one report. Now they can maybe be freed up to. We don't have this piece of information and they can make that human connection with the person who hasn't pooled their data or like they're freed up to troubleshoot while the system does the labor intensive piece. Chris: Exactly. So the AI agent will see that A POD isn't there a piece of document isn't there. It'll follow defined steps. It'll reach out. Write an email that is seemingly coming from a human. Yeah, that that looks like OK, someone's asking for something. And which happens usually today with people, they'll get a response. They'll take that attachment. They'll go upload it somewhere, so I think that's just a great, you know, area of focus and that's something we've done. Jeff: So I guess again putting myself back in the position of I'm a supply chain executive who sees this is an incredible tool that is coming into market. What does getting started look like? What is a realistic timeline to like Try a pilot. Make sure that it is actually you know the right data going in and the right output coming out because they're still going to be a person checking that for a long while to make sure you know it's representing your company. Chris: Yeah. I mean, you're gonna have to have some technical teams that get started on this, that choose the right technology and then confine the data that it has access to and deploy it within your ecosystem. So it's not like a open source out there, you know, able to feed proprietary information to that to that agent. So that you really confined it. So starting down that journey, you want to start looking at prompt engineering. A lot of this coding, which is really exciting, can be done by non technical people through training and learnings and we've used a few technologies including Lang Chain which helps us kind of chain together smaller tasks into an overall complex workflow as well as open AI. And then we have to proprietary models that we've built within the logistics space to then give it data as well as our system of record to have access to information to go after that. So setting up that infrastructure probably takes, you know, two to three months to get started, ensuring you're checking down with your data security data privacy teams to make sure that they've signed off on it and then really starting on very small use cases, learning as you go and then figuring out how you can scale it and what we're doing to scale is basically allowing a planning large language model to look at SOP's that humans are writing. So I can basically write an SOP for a specific action that I want to automate. The planning LLM can learn from that SOP, then define tasks that smaller, specific LLM will follow. It can reorder the workflow based on what it's learning and then have that tangible outcome of, you know what a human would have done in the past. Jeff: I'd like to get into what makes a TMC a partner of choice in this space. I understand you're an early innovator and you've got some experience which already sets you apart from a lot of people, but walk me through how you got into this, what specifically you're doing that is exciting. Some supply chain leaders like walk me through that. Chris: Yeah, I mean one, we wanted to share what we're doing just to show the opportunity in this space. And then I think specifically, why to partner with us in this in this arena is that you know, we probably have more access to transportation data than really any company in the world. So we have $30 billion of freight that's on our platform. So creating very logistics focused models is where we're gonna have that, you know, IP or intellectual property that companies can plug into. So, you know, I think we are a great partner for that, but we also want to be consultative and just talk about the true and tremendous opportunity that this type of technology can bring. And definitely we are very early on, but we're really excited about some of the you know, logistics specific models that we're building, some of the capabilities that we've already had from an engineering and talent development portion that are now focusing really on this technology. And we're vesting a lot more into this space. So we really got the ball rolling trying to get ahead of, you know what opportunities are out there. So I think you know companies that are looking to partner especially in the transportation execution phase should look at TMC and C.H. Robinson as vital partners. Jeff: Absolutely. And you know, not to put words in anyone's mouth. But again, if I were a supply chain executive says I am interested in this, what are some of the questions I should be coming to you with? Like I think there has to be a challenge to the business that needs solving or like what are some of the right ways to approach this rather than just. I want to try the new toy. Chris: Yeah. I mean, people do want to try the new toy, right? It's very it's in the news. It's in the press, so I think first is stack ranked the opportunities create the business case you're going to have to have development, you're going to have to do things on your side. It's not completely something that you can outsource to a partner, though. There are some scenarios that you can find it, but I've really the informed customers to take the time to create the business case, find areas within their supply chain that this could be a focus technology and where maybe they have blind spots in terms of logistics data visibility documentation that we mentioned and discussed earlier. Then looking at your partner network and seeing who's actually investing in the space. So that could be used that could be others. But really, stack ranking, your opportunities being very concise and constrained on what you want to go after, and then partners like us can really hit the ground running with that as a as a you know guardrail to development. Jeff: OK, so if I've got my pain point or opportunity that I'm saying, I think this is a good place to start and I come to C.H. Robinson. And I thank you for TMC. The division of C.H. Robinson that is working specifically in this space, what does day one look like? What does the actual beginning of that partnership look like, and how do I prepare my team to get started? Chris: Yeah, I think one, we want to look at the data model. So make sure that you know we use our proprietary platform Navisphere. So we'd wanna hook into your logistics network. You want to get some data. You really do need a large set of data to get started. If there's blind spots in the data, the AI agents gonna struggle to answer questions or be concise. So I think looking at the data model, seeing where you know we can integrate and support and then you know deploying that business case and going after specific scenario test and learn fail fast you know and fail cheaply if you go after something that's maybe too robust you don't want to take too much time. So getting started very succinctly, moving fast and identifying the data model that you wanna use to go after that is incredibly important. It allows you to move fast in the space. Jeff: Whenever we talk about a new technology, especially something like AI that is very buzzy. Yeah, I do want to take a minute and talk about the people. Part of all this because there's an existing supply chain organization. Yeah, some of what we're talking about is automating their jobs, which will free them up to do more important work. But there is a struggle there for a minute. Can we talk about that? Chris: It's the human equation, right? As a part of it. So what I would say is like our industry in general has a talent gap. So we are struggling to put logistics expertise, transportation expertise, even supply chain expertise in seats to scale out organizations as we grow even you know through this tumultuous time that we've been through, there's still a lack of expertise. So it's not about replacing human capital, in my opinion, it's about maximizing it and being able to scale across many different flows and not having to go higher and go extend your team and you can keep it really focused team focused on the value added initiatives. Jeff: We've covered a lot of ground in this conversation. If there were two or three key points that you want people to think about a little further, yeah, what would those be? Chris: Yeah. So we're very early, it's very exciting, but we got to be very careful and very deliberate about what we're doing with this technology. So a lot of learnings are to be had. We want to make sure it's controlled and we were very rigid with what we're allowing it to do. I think we want to focus on the human element. I think we want to make people's lives better and not necessarily, you know, focus on replacement, but augmentation of things that you know people don't want to do, we want them to focus on the cool things that really bring value to their companies. And this is an Avenue to do that. So starting small learning quickly, assessing your partner network on who actually is investing in this type of technology, who you can partner with and then looking at your own data model to start. Jeff: I have to think with something this new and this exciting, they're going to be people with questions, people want to learn more. What is the best way to get in touch? Maybe pick someone's brain at TMC. Chris: Yeah. So I would direct people to go to thechrobinson.com/TMC and there's an option there to find an expert. So you can learn more about what we're doing in the space happily also be consultative and what we're doing and how we've how we've done it. So you can actually build that out on your side and potentially we can find a way to partner in the future. Jeff: I really appreciate your time here, Chris. This this is a really exciting thing and I actually haven't gotten to talk about it with someone who's really doing it yet. So this this was exciting for me. I really appreciate you being here. Chris: Thank you and thanks for executive platforms. You guys run great conferences, had a great time here and look forward to more in the future. Jeff: Fantastic. In the meantime, you've been listening to another episode of executive Platforms Blueprint podcast series. I've been Jeff Nicks. Let's do it again soon.
Supply chains are inherently filled with people and manual, repetitive tasks. New technology, like generative AI, can automate the required activities so more time can be spent on the those that add value.
Unlike traditional AI, which is narrow, task-oriented computer intelligence, generative AI is broader, human-like intelligence capable of learning and adapting. This makes it the ideal choice for certain areas of the supply chain. Imagine the possibilities this could have on your own strategy.
Chris Cutshaw, director of market solutions at TMC recently joined Geoff Micks on the Executive Platforms Blueprint podcast to discuss the role and potential impact of generative AI on the transportation and logistics industry. As part of our focus on applying generative AI to benefit our shippers and carriers, this video offer a quick view of how C.H. Robinson and TMC are helping inform the future of supply chains.