Toyota's Secret AI Playbook

Chad Watt: Welcome to the Infosys Knowledge Institute podcast, where business leaders share what they've learned on their technology journey. I'm Chad Watt, Infosys Knowledge Institute researcher and writer. Today, I'm speaking with Jason Ballard, Vice President of Digital Innovations at Toyota Motors North America, and Hanuman Gadepalli, Associate Vice President in Automotive at Infosys. Welcome, gentlemen. Jason and Hanuman are experts in supply chain. Today, we will discuss artificial intelligence and how it can be deployed in the automotive supply chain. Thanks for being here.

Jason Ballard / Hanuman Gadepalli: Thanks for having us.

Chad Watt: Jason, give us some sense of the size and scope of Toyota Motors North America's supply chain.

Jason Ballard: Yeah, it's a rather complex supply chain. It's one that has evolved over the decades. It was started as a vehicle supply chain, and then we introduced it as a parts supply chain. We've been on a mission to form that one Toyota supply chain. In North America, our growth has been substantial. We're fortunate enough to sell 2.4 million vehicles every year in North America alone. We've got a great dealer network. We believe the growth is up to around almost 1,600 dealers between Toyota and Lexus across North America. So it's amazing the orchestration that happens across North America to get the parts from our tiered suppliers into our production facilities, all the way routed to all of our dealerships all across North America.

Chad Watt: Now at a high level, how does Toyota think about artificial intelligence?

Jason Ballard: Yeah, at a high level, we're very bullish on artificial intelligence. As we think about how to reimagine our future and how we reimagine operations happening inside the company, it's going to include emerging technology, and AI is certainly at the forefront right now. We've been leaning into it over the last several years. When AI used to be referred to as machine learning and data science, now it's agentic. Who knows what it's going to continue to evolve into? But we really believe we're not using technology for technology's sake, but we're using technology where it's necessary to evolve our processes and drive our operations forward, thinking about our team members, our customers, and the company overall.

Chad Watt: Great. Now, in the context of supply chain and AI, what typically blocks AI pilots from reaching enterprise scale?

Jason Ballard: That's a very popular question. You read a lot about that, and certainly we've experienced those challenges internally. The analogy I like to compare it to is, do you play golf?

Chad Watt: No. No?

Jason Ballard: Well, when you're on the range, you're freely hitting these golf balls. You're like, I can't be stopped. Right? The ball's coming off the club so well.

Chad Watt: Yes.

Jason Ballard: You tee it up on the first tee, you got the trees lined on both sides. You got wind in your face, right? And it's a lot more difficult. There's a lot more variables. Usually, the AI pilots, you're in a very tight, isolated environment, and you're specifically focused on one or two steps in that process. But when you try to take that pilot and extend it into production, you're introducing a lot more variables. One, your data. Nothing is going to be successful with AI unless you have production-ready data and context is understood. Integrations are super complex. In an environment like Toyota, where we've been growing over the last several decades, we've got various types of systems. I'm talking spreadsheets, mainframe, distributed systems. And so the integrations with that are complex. And then once you include all of those facets, then you really have a focus on the return on investment. And sometimes what you thought you could achieve, that hypothesis in the beginning, doesn't necessarily pan out unless you truly take a wide view on how to incorporate all of that from the beginning.

Chad Watt: Great. Hanuman, can you add to that? How do you get from pilot to scale?

Hanuman Gadepalli: So what we are seeing is pilots are nothing but a proof of value. The way, I think, Jason, you articulated. The complexity of the data, the integrations, right? These are all very critical.

Jason Ballard: Yeah.

Hanuman Gadepalli: At proof of value stage, you just see how a piece of stuff is working. But to put it in production, it requires huge amount of work. So that is where I think we are embedding AI into the business decision loop, where we are making sure that the AI and the governance is well plugged in so that you can realize that as a platform.

Chad Watt: Jason, I want to come back to the supply chain professionals. How are you involving supply chain professionals, not just developers, in building and adopting AI?

Jason Ballard: So I don't think you can achieve the outcomes that you want if you're not working side by side with your operational experts. I'm fortunate to lead digital innovations. We're a blended team of operational experts and technologists, and we sit in the supply chain business within Toyota Motor North America. And so that close proximity, us truly understanding their problems, the operational folks understanding more and more every day about how the technology works, it really creates a great one-two combination because they know how to rethink the operations, especially those forward-leaning operational team members, and those are the ones that you have engaged in the transformation. But sometimes they don't know exactly how far the technology can enable them to take that operation. So it's a give and take, sharing information back and forth with one another, and then that's exactly how we take these AI use cases forward, in conjunction together as one team operating together.

Chad Watt: Great. Hanuman, you want to add to that?

Hanuman Gadepalli: I think what Jason said is, the one team is really the one where how it is coming across in AI, right? There are people who know the domain and the context. They are the SMEs. So there are developers who are making the AI technology work. Both these work together, where that is what we call it as domain in the loop AI. So, we believe in co-creation, where the planners or the real SMEs come together and validate the models which developers are preparing. So that is where it becomes more one team working on to realize the AI value for that specific problem, which we are addressing through AI.

Chad Watt: Great. Jason, come back to this and ask you to kind of put a finer point on it. What proportion of your AI work is solely in the hands of developers, versus in a kind of a hybrid group of people?

Jason Ballard: I would say, putting a number on it, what we've been doing is we've been focusing on how to, within my group, and across Toyota Motor North American in general, is how to create more builders in our organization. Right? We want to have more ownership of the future that we're building. But we can't do that without partners like Infosys. And so, there's definitely certain roles that we are completely dependent on Infosys and other partners for, right? There's certain roles we want to own, but use partners for scale. So I'd say we're still pretty lean on the team members, but that is an area where we're focusing more and more on to grow and extend, because that's truly how we think about the flywheel and the acceleration of transformation is by taking more ownership internally.

Chad Watt: Describe how you were thinking about your AI model strategy going forward. Are you standardizing on one single ecosystem or building kind of a multi-model AI architecture?

Jason Ballard: Yeah, it's definitely the latter. We have to... I think right now everybody should be doing that. I'm not sure. I mean, just but speaking for Toyota, I mean, we're very focused on, these models are evolving and changing and leapfrogging one another. And so, certain models are really great at certain types of use cases, and we think about the breadth and depth of our supply chain, I think we want that capability to look at and leverage different models, based on the types of use cases we're trying to solve for. So something in the demand forecasting, demand planning realm may require one type of model. But how we're doing maybe customer front-end intelligence, understanding the true customer needs, may require something differently. So we're keeping our models in the ecosystem pretty open right now, as things are still growing in this space.

Chad Watt: Hanuman, what do you think about a multi-model strategy?

Hanuman Gadepalli: I think what Jason said is the right direction. I believe, strongly believe in that, because the model adoption is going to be a dynamic, and future is going to be a model marketplace. So we strongly believe in having these orchestration layers latching dynamically onto the model, which gives best optimization, best performance, best fit for the use case. So that's what I think it's going to be the direction going forward.

Chad Watt: Terrific. Jason, when do you expect AI to start making autonomous decisions in your supply chain? And what's got to happen for you to be comfortable with that? Do you have a crystal ball?

Jason Ballard: I don't. But I love the question. From Toyota's standpoint, we partner with various other companies. One partnership we have is a company called Zero100. They have a focus on digital supply chain, so they're doing a lot of research, sharing that research with customers like us, and they look at the entire loop of the supply chain, and they break that down into various types of use cases, and start to share already where they're seeing other companies investing and where you might just see your processes become more assisted processes all the way through more autonomous processes. So I think when you think about the end-to-end operations, the end-to-end workflows, there will be some opportunities for some autonomous workflows to occur. We're not there yet. I don't want to make it look like we are, but we've got this framework called Decision Agency Framework, and it really leverages Toyota Production System. Everything we're doing in our transformation runs on the foundation of TPS. We never want to forget that. And it's really crucial for how we go forward. But this Decision Agency Framework is really about how do we think about version one, which may be a more aided, assisted manner, all the way through version four, which could become more of an autonomous, where it's acting independently. But at the end of the day, though, we're really focused on that the team member must have a significant stake in final decisions, right? And again, it's not saying they need to make every decision, but there will be that criticality that our team members are involved. This isn't a black box operation. That we've got a confidence gate, that the AI is performing as it should, and then we have a learning gate that the team member is understanding how it's working and feels comfortable to keep evolving the models. So that's kind of how we're seeing that journey from assisted to autonomous.

Chad Watt: Got it. Assisted to autonomous. Hanuman, what's got to happen, and how do you approach this?

Hanuman Gadepalli: I would say, I think in line with what Jason said, I think it's a journey, a maturity journey, I would call it as. Where it goes a phase-wised approach, where you go to the assist and then recommend and then autonomous. So that means it goes stage by stage. And it comes with the confidence, building the confidence with explainability that comes with how you are actually driving the assist to recommend to automate, with governance, and then giving the trust and confidence to the stakeholders. That defines the autonomous scale.

Chad Watt: I'm going to come back to something you touched on earlier. Is your data ready for use with AI and how do you get it ready?

Jason Ballard: Of course, it is. Talking about the complexity of our supply chain, I will say what we did when we started the transformation several years ago is we put a focus on a platform first. So internally, we refer to this as Cube, but it's a platform where it's got various layers to it. At the foundation, the bottom layer is the data core. So it's like where we bring together our vehicle hub data, our parts hub, our logistics pricing. And so we've been working on that for over four years to get that data collated on the platform so that it's handling all of our operations. And then there's various layers on top of that. Right now, we're building out intelligence layers. So that includes semantic layer, knowledge graphs, really providing context to the data that it's integrated with, that it's connected to. That we're still evolving and maturing. So that's where we didn't create the platform first. We allowed the value-driven use cases, like what we wanted to transform first, to create the platform. And so we've been not thinking we have to address everything at once. It's been more of step by step, and so mature, bring in new data, mature that. And so, that's been our approach to how to get there. We're confident and bullish we will, and we know that that's going to be super critical for how we eventually get to having a multi-agent orchestration happening across our supply chain is when we get that in place.

Chad Watt: And to that, Hanuman, what does it take to get data ready for real... to really derive value from AI?

Hanuman Gadepalli: I think data is foundation. Most of AI successes depend on how the data is organized. In this case, the way I think Jason talked about the criticality of the vehicle data, it is about how we are standardizing the data, and how we are defining the semantic layers, and how we are defining the governance of the data, is going to make sure that your data is ready to implement AI on that data.

Chad Watt: Jason, the Toyota Global Program plans to integrate Toyota supply chains across continents. How is AI applied to this initiative?

Jason Ballard: Yeah, so maybe a little context. So again, Toyota is a global company. We operate multinational, where regions have a lot of autonomy. But we still have some global systems with our headquarters, Toyota Motor Corporation, have created over the years. And typically, when we create global systems, TMC is the one to do that. But through the work that we've done in North America, they saw it as a great opportunity for us to co-create and partner on this, what we're calling Project Onward. And it's really looking at the foundation of what we've delivered in North America over the last few years, the platform, pipeline management, advancements and demand in supply, matching capabilities. And now we've got the support, and we're building out a global platform, a global pipeline management system, where it's every vehicle record for every vehicle across the globe is in this system, and you can see where it is in the production process all the way through delivery to the dealer or delivery to the customer, depending on the region's operations. So it's very powerful, and it's a great opportunity. It's one that I've been with the company 23 years, and I can say it's probably the first where we're truly co-creating with TMC on this. So it's a once in a career type of opportunity. And so the team is super, super zealous about what we're able to create here. As far as an AI perspective, again, we're leveraging all types. So when we think about demand forecasting, it's data science, machine learning models, how do we do better optimization engines. We're looking at just different ways that we get the customer truth, like what they're truly interested in, building into these forecast models so we can be more accurate. And we make sure that we've got the right vehicle, the right place at the right time for our customers. But then when it comes to pipeline management, this is where we're also looking at scenarios where we include agentic capabilities. Because in the pipeline is where we track things like estimated time of arrival. We can see when the vehicles are supposed to be delivered versus maybe when we initially communicated the target delivery date. So there's ways that we're looking at agentic to help keep those vehicles on track, to inform of delays to our logistics providers or our dealers, depending on the situation of what's being handled. So we're using AI in various forms and again, being very bullish on leaning into it because we really feel like this can be a true disruptor for how we run global supply chain.

Chad Watt: Hanuman, you've got a global company that does things very well in different regions, coming together with AI as the tool to link things in new ways. What have you noticed about this?

Hanuman Gadepalli: I think in this case, I would look at AI as sort of like a unifier here, where how you define data models or intelligence layers across geographies, taking care of privacy, security, compliance, which is specific to that geography. AI is going to play a critical role in these kind of global rollouts across geographies.

Chad Watt: So Jason, how does the Toyota Production System influence AI adoption?

Jason Ballard: I'm glad you asked that question. Toyota Production System, again, is at the heart of everything we do inside the company. Most people when they first think of TPS, they think of Kaizen, Jidoka, pulling the Andon, and those are certainly elements. But when you take a step back and look at the broader definition, it's really about empowering our people to create time for them to work and solve the most critical, innovative problems. The foundation of this was established that you go to work to think. And so everything we're doing from a transformation standpoint is like, how are we able to free up our team members so that they're there to think and are empowered to solve the most critical challenges that they're dealing with every single day. So, as we think about reimagining our operations, think about how we layer on AI, TPS remains foundational to everything we do.

Chad Watt: Jason, Toyota's bullish on AI. Some folks are a little more intimidated by it. How do you manage this kind of big change that employees and society is facing?

Jason Ballard: Yeah. Well, I think, people read a lot, and you hear a lot and, right now some of the stories are more like, if I'm an employee, I get a little bit worried, right? Because it's like AI is going to eliminate jobs, right? And so in Toyota, one of the things that if you go back to when we've started, we rarely ever lay off anyone. And that was early in Toyota when that happened. And so we're very committed to our team members, and we have a lot of work that needs to happen inside of our company. So the way we look at it, if we can take a process, a workflow, and we can automate it or, and eventually provide autonomous capabilities, that doesn't mean the work for that team member and the team member needs to leave, right? It means the team member now gets to go take on some higher valued activities or maybe move to a different part of the company. That's another thing within Toyota. We really stretch our team members. We try to challenge them. Growth comes through rotation. So that's kind of how we're approaching it. But it's not easy to get that message out to everyone, and still people want to, they need to see it to believe it. So it starts with leadership, just making sure that that messaging is clear. Being a chief reminding officer and finding ways to story tell. And use examples of things that we have done and how it has uplifted the team member. I mean, just for context again, we're building a new global planning system, that we refer to as GPS. It's replacing the team members having to manually operate 75 different spreadsheets to perform a monthly task. So think about that. As far as the complexity of what they're doing in spreadsheets, now they have an intuitive system with AI incorporated into it. So we don't take as many people now to perform that. So now we can shift those folks to some other, again, other parts of our supply chain where the need is, and their thinking way is necessary. So that's how we view it.

Chad Watt: Wow. Terrific. Hanuman, how do you prepare workers for AI?

Hanuman Gadepalli: I think this is across enterprises. Like the adoption of AI, I would say it creates two categories of transformations. One is technology transformation. The other one is cultural transformation. That's what I think. Jason explained it beautifully. So people, if they adopt AI, they will have more ways to spend their time in innovative tasks, and then take the advantage of AI. It is nothing to worry about, but it is more about how you are adopting and how you are using it for your advantage.

Chad Watt: Great. Hanuman, Jason, thank you for your time today.

Jason Ballard / Hanuman Gadepalli: Yeah. Thank you, Chad. Thanks, Chad.

Chad Watt: This podcast is part of our collaboration with MIT Technology Review. Visit our Enterprise AI hub on technologyreview.com to learn more. Be sure to follow the Infosys Knowledge Institute podcast wherever you get your podcasts. You can find more details in our show notes and transcripts at infosys.com/IKI in our podcast section. Thanks to our producers, Christine Calhoun and Yulia De Bari. Dode Bigley is our audio technician, and I'm Chad Watt with the Infosys Knowledge Institute. Until next time, keep learning and keep sharing.