Baxter has faced multiple supply chain challenges over the past year. Hurricane Helene took its North Carolina IV fluid plant offline, causing months of disruption, and at the same time the company was dealing with "unnaturally high order volumes" for blood pressure cuffs.
Even before the floods and raw material shortages, the company was actively seeking supply chain optimization. A.K. Karan, global vice president of advanced engineering and innovation at Baxter, has been at the forefront of the company’s efforts to use artificial intelligence and machine learning to become more proactive in addressing natural disasters, political instability, supplier problems, global health crisis, labor shortage and other potential roadblocks to consistent production.
After his keynote address at MD&M East, Medical Design & Development sat down with Karan to discuss his company’s approach to using AI and ML not as magic remedies but as tools for addressing pain points throughout the process.
The following transcript has been edited for clarity and length.
Medical Design & Development: You joined Baxter in 2021. How far along was the company at that point in terms of machine learning and AI?
A.K. Karan: In the manufacturing setting, it was very minimal. We tried a couple of use cases and this was before my time. We didn't see results so there was no future activity because we were not seeing the value.
MDD: Was there one particular implementation that really delivered that first good chunk of value?
Karan: Once I started with Baxter, we saw the value in very challenging use cases, especially in using AI vision for product inspection. We were working with one of our plants using a dialyzer fiber. We had to look for certain patterns and this thing runs at a very high speed. We wanted to see if we could bring machine learning and AI to cut inspection time. because this is something a human cannot do. If a human can do it well and good, that’s fine, but we wanted to find the most challenging use case as a proof of value.
We did that project in less than a month. That's when the leadership says, 'You know what? This is something that we thought was not possible, but this team was able to do it within a month.'
Machine learning doesn't have to be complex. If you use the right tool, it can be very fast. And working with the right partners, you're able to have a value proposition in a matter of weeks.
But you have to understand the pain point. What am I trying to solve? Can I bring the right tools to solve this problem? Then you prove it and you take it to all your global sites. The upfront investment could be a bit on the higher side when you go in, but once you start scaling it, that's where the price goes down and the value goes up.
MDD: You spoke about value in terms of getting that first win with these implementations. How important is that win in terms of ROI; establishing that something can at least produce value before you find new use cases?
Karan: When you talk about point solutions, you're talking about this particular inspection. That's one, right? We also talked about predictive maintenance. They are very purpose-built solutions. They do a job, a good job, a great job, and then boom, you can scale. Let's shift gears.
Right now we are going into the large language models or the ChatGPTs of the world, generative AI. People say you can't quantify these models yet, because it's hard to put a value to it.
For certain tools, you have to build it and then they'll come, but you have to be mindful of the investments. If you build strategic partnership, then it's a co-investment instead of us just pouring in several millions of dollars.
When you're working on the bleeding edge, there's always a financial risk, but if you get a home run, then that thing can just grow exponentially.
MDD: How do you find a strategic partner for something like that? Is this something where you have to have a mutual benefit?
Karan: Exactly. We have partnerships with some of the leading players, Amazon Web Services (AWS) being one. We do a lot of mutual developments and they have invested money in some of our projects, too. Because if you look at AWS, they do have what is called the Gen AI Innovation Council. They invested like $100 million; this is public info.
They wanted to work with one of the leading companies that’s actually breaking the mold and working on the bleeding edge. So, they worked with the Baxter manufacturing team because they knew how we are building and deploying solutions. They're investing in us to build a solution for our needs, because when we start scaling, it’s a win-win for our strategic relationship.
MDD: You joined Baxter at the tail end of COVID, which was a huge disruption. But you've also seen some other pretty significant disruptions in production at Baxter, most recently at the North Cove facility after Hurricane Helene. Can you talk about the work you're doing and how it connects back to something like that?
Karan: My team is also responsible for building our automation, digital transformation, and operation excellence. If you look at Hurricane Helene, that's a 200-year-flood situation. No one expected that would happen, but it happened.
You can have the best of the best tools, but still that isn't going to be good enough. This is pure grit, sheer determination, and all the perseverance of the entire Baxter workforce. The team united like never before. I’ve never seen a company come together globally like that. At the end of the day, it’s what makes me get out of bed. Technology is great, but we are doing things to save and sustain lives and that's what makes the whole Baxter team tick.
However, technology did help us get the machines back up and running. How do you know the machine is still operating in the same condition? That data becomes key to make sure it's all working in good condition.
All of the data is in the cloud. We had some data center breach because of the flood, but in this case, we are much safer because the data resides in the cloud. We can show our team how the machine ran before the hurricane.
MDD: I want to shift topics to workforce management. I think you called it 'upscaling workforce,' but you showed a VR training application. What role did AI and machine learning actually play in development?
Karan: You’re creating a 3D asset in your virtual world and you are providing the experience to the user. But the process of creating that asset itself is quite complex. It's like creating a video game. So where AI comes in, it helps with the workflow—streamlining the workflow as you are building the module. Before that, one of those modules would have taken at least a year to build because every single thing has to be created manually.
But now that is getting compressed because you have more workflow automation tools with AI coming in. Instead of taking one year to create that asset, you can create that asset in a matter of weeks. Once it gets deployed, it's much easier for the user to learn the machine. I keep telling people to break the machine virtually, because the more you break, the more mistakes you make, the better you get. Then, in the real world, you won't be making those mistakes.
MDD: It's like a gamification of the training process, and you said that this is something that's become a little bit more expected with newer workforces. Are all these use cases for AI and machine learning becoming a little bit more accepted within the workforce at Baxter.
Karan: I think we have to be very careful when people say, 'I'm going to use AI and ML for everything.' It's not a silver bullet, and it also comes with its own caveats. It's learning from the historical data and patterns. So, how good your data governance is and how good your data is directly relates to how good your AI is going to be. We feed garbage in, we get garbage out.
We are very cognizant of the fact that we use the systems where we can show value. I mean there are certain instances where we tried AI and very quickly we found out this is not going to work.
MDD: Do you have an example of a situation where Baxter tried to leverage AI as a solution for a pain point and it didn't really materialize?
Karan: When I started with Baxter, it was as a plant that is not in existence anymore. But it was a very, very complex, very intricate process. It was just pure magic how we used to make these dialyzer fibers.
On one of the machines, you have like 240 fibers running at 55, 60 meters per minute. And they're so delicate, the fiber would snap if you blew on it. It would also take almost eight to 10 hours to reboot the machine. It was causing a lot of pain points so we were trying to use AI to predict the fiber's behavior.
We spent around three months with one of the big players in this space. I don’t want to name the vendor, but it’s pretty big in this space. We couldn't find any anomalies. So we said, okay, maybe this isn't the right time to embark on this use case and we stopped it pretty quickly.
MDD: Where do you see the future of optimization coming from in terms of use of AI and machine learning?
Karan: When you talk about supply chain optimization, you're looking at how can I make products of the highest quality at the lowest cost. That's optimization in very minimalistic terms. You're constantly looking into how I can optimize things, because in the manufacturing process you have to make the production at the right time, the right amount, and also minimize your quality issues.
Automation is great, but you still need a human to do certain things at the right time.
