Thanks to AI, we gained control and prediction. Our investment paid off within four months, says Artur Heider from Hyundai

April 15, 2025

Terms like AI, optimization, and efficiency are everywhere. But what do they actually mean in practice? What does production optimization involve? And how do you tell a buzzword from a real solution? Learn more in our discussion with Artur Heider, EM Production Planning Specialist at Hyundai.

Read the podcast as an interview: 

(The interview has been shortened and edited using ChatGPT.) 

Ivana Karhanová: In partnership with Adastra, you tackled two specific use cases at Hyundai. One was the welding shop, which has 13 constraints in daily production sequencing, and the other was the paint shop, where the goal was to reduce material consumption. What do these constraints in the welding shop look like in practice? 

Artur Heider: It's not easy to explain for someone who's not familiar with the process. In the welding shop, we have around 300 robots and devices that need to be fully synchronized—every cycle must follow the next precisely, and everything has to be in the right place at the right time. It's like a giant clockwork that has to run perfectly. 

Welding is the bottleneck in our entire production process. We struggle with space and other challenges there, so we need everything to fit and function exactly right. For example, on one line we produce several car models, so we need to load the line efficiently. We can't just build one model all shift and then another the next—we need to schedule them according to takt time. Say, the Tucson model must appear with a specific spacing, the same goes for the Kona Electric and i30. That means we can't just run all Tucsons in one shift and all i30s in the next; we need to balance them so that the machines can handle it and we get the best output from the line. 

Ivana Karhanová: And what about the paint shop? 

Artur Heider: That's a bit different. After every change in exterior colour, the robot that sprays paint needs to be flushed with thinner and other materials. Each colour change costs us money because we need to clean the equipment, and it also takes time. 

Ivana Karhanová: And the painted bodies coming out of the paint shop have to be coordinated with the rest of production, right? So you can't just paint one colour all day. 

Artur Heider: Exactly. The line works continuously, and we don't have to alternate the colours in a specific order, but it's best for us to run the same colour as long as possible to reduce flushing cycles and save material. 

Ivana Karhanová: When did you realize the process needed optimization? What triggered that thought? 

Artur Heider: From the beginning, it was never just about manual planning—our planning process runs through systems that were put in place when the factory was first built. The initial impulse came from the paint shop team. They knew that thinner and cleaning material usage fluctuates depending on how frequently we switch colours. 

We did have some software for production sequencing, but it didn't account for all the constraints we have in the welding shop, and it couldn't properly optimize the colour sequence for painting. So we knew the process could be improved, and we already had a pretty clear idea of how the new software should work and what exactly it should do. 

Ivana Karhanová: When you decided to go for a proof of concept—you tested it first before deploying it into production. What did you define as the success criteria? 

Artur Heider: That was actually quite simple. In the PoC, we built the optimization model and tested it using real production data. The output was something we could feed into actual production, and we ran that for two or three weeks. That way, we could directly measure how much material we saved. The paint shop confirmed that the solution would bring significant long-term savings. 

Ivana Karhanová: Can we reveal how the PoC turned out—what was the ROI and how big were the savings? 

Artur Heider: We calculated the savings to be around CZK 13 million per year. I think that's a very strong result. The return on investment was about three to four months. 

Ivana Karhanová: Optimization algorithms that use machine learning are highly dependent on data quality. How were you positioned at Hyundai when it comes to data readiness? 

Artur Heider: I think we didn't have any big issues with data quality. As a large corporation, we use various databases and data is stored well. The problem—common to many big companies—is hesitation to invest in new systems and technologies because change is hard. 

As for connecting the application, I don't think it was a big issue either. Of course, there are IT factors like firewalls to deal with, but overall, it went smoothly. The application connects via API to our system, downloads the daily production plan, optimizes it, and sends it back. It's a side process that runs well. 

Ivana Karhanová: You mentioned that deciding whether to implement such a tool, platform, or software isn't just about finances. Did you face internal resistance or mindset shifts? What made it complex from your perspective? 

Artur Heider: The biggest challenge was the corporate mindset. Hyundai-Kia follows a philosophy that everything must be applicable across all plants—whether in Korea or Europe—so they all operate on the same principles and software. In theory, that's good for maintenance and consistency. But in practice, each plant is very different—different models, suppliers, material sourcing—there are many variables. So software needs to adapt to local requirements. 

We'd had the idea to optimize sequencing for several years, but it wasn't approved by leadership. Over time, more decision-making power was given to the Czech plant. We showed that Nošovice can innovate, bring in good ideas, and save money. That made it easier to finally get approval and implement the project. 

Ivana Karhanová: Do other plants now use the sequence optimization? 

Artur Heider: Yes. We went live at the end of last year, so we've been running it in production for about two or three months. The Kia plant in Žilina—our sister plant—heard about it, came to see it, and is now going through their internal approval process to implement something similar. Colleagues in Turkey saw it too. Once word gets out that something saves money, other plants become interested. 

Ivana Karhanová: And you even received an award from Hyundai for this—don't be so modest. 

Artur Heider: Yes, I did get recognized. But for me, it's mostly about enjoying the work. I like thinking outside the box and figuring out how to make work simpler and more efficient. 

Ivana Karhanová: Was there anything that surprised you during the PoC? 

Artur Heider: Yes. Even small changes in how the sequence is planned can impact suppliers. For example, one car model with a specific paint colour that gets ordered frequently ended up clustered in a short timeframe, and our local suppliers had to suddenly ramp up deliveries to meet our just-in-time and just-in-sequence requirements. They weren't fully ready. We hadn't expected this kind of ripple effect, but it was resolved in a week or two—suppliers adapted quickly. So now, there are no issues. 

Ivana Karhanová: Looking back, what were the most important lessons from this optimization? 

Artur Heider: Definitely, to not underestimate preparation—defining the actual problem is key. Many people have vague ideas about what could be improved, or they want to jump into AI without knowing how it works or what it requires. 

You need to define the data you have, where it is, and what exact number you want to improve. Sketch out how the process should work and what constraints exist. In our case, I'd already been working on this topic for a while and had even developed a simple optimization model myself. So I could present Adastra with a fairly well-formed concept. That made things easier. 

Ivana Karhanová: How long did it take you personally to prepare all the systems and materials so the optimization could be applied? 

Artur Heider: That's hard to say exactly. I've been in production for about 17 years, and we deal with planning daily—short-term and long-term. But I'd say preparing everything—agreeing on how to collect data, mapping constraints in the welding and paint shops, speaking to those teams—can be done in about a month. 

Sometimes, during testing, new constraints come up that no one mentioned at first. So things get refined along the way. But from start to a working concept, I'd say it can be done in about six months. 

Ivana Karhanová: Now I understand why you got that award. Preparing the data in a month and going live within half a year—that's impressive. Do you see more potential use cases now? 

Artur Heider: Yes, definitely. Every company wants to optimize and save, so people bring up various ideas. But not all of them are worth it. 

We actually worked with Adastra on another PoC for optimizing container loading. But, during the business analysis, we realized that while optimization was possible, the ROI was much lower than for sequencing. So we decided not to proceed. I think that's fine—it's better to analyze first and have an expert tell you whether it's worth pursuing. 

Ivana Karhanová: Do you have an internal system for collecting optimization ideas? 

Artur Heider: Yes, the process works on several levels. The impulse to optimize comes from top management—they set cost-cutting targets. That motivates people to think about improvements. There's also positive motivation—good ideas are rewarded. The ideas are collected and evaluated to see if they're worth the investment. 

Ivana Karhanová: From your personal view, where should attention go next—what has the most potential for reducing costs? 

Artur Heider: I'd say data management in logistics and planning. We're still using legacy systems, and employees don't have easy access to data. Many still use Excel macros to pull data from SAP and upload it somewhere else. 

That's hard to quantify in terms of saved man-days or costs, but better data access would let us track trends, apply machine learning, and later AI. That would definitely lead to savings. 

Ivana Karhanová: So you're saying we should also pursue innovations or use cases that don't immediately deliver ROI, but lay the groundwork for future impact? 

Artur Heider: Exactly. Many companies want AI without having their data ready. They bring in someone to build a chatbot or something else, but if the data isn't in order and users can't easily access the info they need, it just causes more frustration. 

Ivana Karhanová: The chatbot becomes another source of frustration. 

Artur Heider: Right. If the data's not right, the chatbot doesn't solve anything. I think larger companies should focus on getting all their data in one place and making it accessible. Once you have that, you can build from there and work with it properly. 

Ivana Karhanová: Those are the priorities in manufacturing, as described by Artur Heider, Production Planning Specialist at Hyundai. Thank you for joining us and sharing your experience. Until next time. 

Artur Heider: Thank you. Goodbye.