‘The First Day Is the Worst Day’: DHL’s Gina Chung on How AI Improves Over Time
24:07 Share

Today, we're airing an episode produced by our friends at the Modern CTO Podcast, who were kind enough to have me on recently as a guest. We talked about the rise of generative AI, what it means to be successful with technology, and some considerations for leaders to think about as they shepherd technology implementation efforts. Find the Modern CTO Podcast on Apple Podcast, Spotify, or wherever you get your podcast. Of course, managers can change processes to use AI.

But how does adopting AI change organizations? AI is a force of change, but change is not easy and it's got to be a learning process. In this episode, Gina Chung at DHL relates how adopting AI can shift a corporate culture to embrace innovation. Welcome to Me, Myself, and AI, a podcast on artificial intelligence and business. Each week we introduce you to someone innovating with AI.

I'm Sam Ransbotham, Professor of Information Systems at Boston College, and I'm also the guest editor for the AI and Business Strategy Big Idea Program at MIT Sloan Management Review.

And I'm Shervin Kodabande, senior partner with BCG, and I co-lead BCG's AI practice in North America. And together, BCG and MIT SMR have been researching AI for four years, interviewing hundreds of practitioners and surveying thousands of companies on what it takes to build and deploy and scale AI capabilities and really transform the way organizations operate.

So in the last couple of episodes, we've talked with Walmart, we've talked with Humana. I'm pretty excited. Today, we're talking with Gina Chung from DHL. Hi, Gina. Welcome to the show. Can you take a minute and introduce yourself and tell us a little bit about your role?

Hi, I'm Gina Chung. I head up innovation for DHL in the Americas region. And as part of this role, I also operate our innovation center out here in Chicago that's focused on helping supply chain leaders leverage technologies like AI, robotics, wearables in our global operations. So how did you get there? How did you end up at that position?

I might actually answer this by starting back in college. So I started college wanting to be an investment banker and very quickly figured out that's not for me. But I took a supply chain course and ended up becoming fascinated by how things get manufactured and how things get distributed.

I think it's something to do with the fact that I'm from New Zealand and grew up in a pretty isolated part of the world. But anyway, after college, I joined DHL at their headquarters in Germany, helped launch eight years ago some of our very first projects working with startups in our operations. And a few years ago, they then asked me to have the pleasure of launching our third innovation center that serves the Americas region out here in Chicago.

Actually, I think we can end right there that way. I love it when someone gets converted from investment banker to supply chain and operations. I think that's great. So can you give us an example of a project that your team has applied a technology like AI to?

Yeah, so one project that we've completed using AI and computer vision is to use it to automate the inspection of pallets in our world. So currently today, our operators have to see whether you can stack one pallet on top of another. And that might seem very trivial, but actually it's sometimes very difficult to identify whether that bottom pallet is going to be damaged. And you have to look for certain markers, certain indications.

And through combining a camera vision system with AI software, we're able to automate that process and reduce potential damages as well as also optimize utilization in our aircraft.

So who uses this system? Who uses it? It's our operations. So people on the shop floor that are helping to load our aircraft, the pallets pass through our system. It flags if the pallet can't be stacked. And then our operators are able to see that and then take that pallet out and give it the right marker to say that it can't be stacked. And then there are some other steps in that process to deal with a pallet that can't be stacked.

Before, somebody would have to be trained on how to identify whether a palette can be stacked or not. So they'd have to be trained on look out for these kinds of markers, these kinds of indentations. And then each palette, as it comes through, you'd have to kind of walk around it and make a note and type it into the system. But now we can actually automate that process using AI and computer vision.

This is a great example of how AI is taking human, unnecessary human role away, probably even increasing the accuracy and precision, I would assume, of like even picking things that human might have missed. Can you comment on how the process that you guys go through to make that AI engine intelligent? I'm assuming there was human involved as it was being designed. Can you comment on how that worked?

Absolutely. So I always say AI is a very broad term. So you can use AI in robotics, you can use AI with computer vision, you can use AI algorithmically. For this particular use case, we worked with a partner, a startup actually.

And together with our operations and a startup, we developed the algorithm for this particular use case. And it was designed with a lot of images initially. So just collecting images and images of palettes, working with our operations to train that algorithm to look out for these specific markers and indentations.

And then after it's deployed, there is a recommendation, right, that this pallet is not stackable. And it's up to also our workers to trust in that. If they think it's inaccurate, they can make a marking and we'll look at it and see if the AI, the algorithm needs to be improved. So we have that loop in the process to continuously train the algorithm because our shipments come in all different shapes and sizes. That is a great point because as we know, without that loop,

it's going to make mistakes and those mistakes will compound. So really interesting how that loop is made. So on day one, when you turn this system on, what kind of reaction did you get from people? How did people feel?

Yeah, we always like to say the first day for AI is the worst day. And what we mean by that is, you know, the algorithm, it only gets more accurate over time as you ingest more and more data and more and more different exceptions. So when we turn on the AI, especially during a pilot, the accuracy looks pretty low and people start to question, hey, I don't think that the AI can actually do this.

But then as we see the pilot go on week after week and it ingests more and more data, it also learns from our workers as well that the accuracy drastically increases and then people start to really believe in it and it starts to make their lives easier. So we try to focus on automating those activities that are really, you know, tedious, repetitive. We try to also build the accuracy to such a confidence level that people trust it and embrace it rather than, you know,

the AI spitting out recommendations that people know aren't correct. Yeah. I think that's a great point, Gina. The idea that it's not being forced as brute force, but the users are involved from the beginning in the design and they actually see it better, their own judgment, right?

I could imagine if you guys did it differently, maybe a more sort of old school way of saying, well, this is the best thing because it's using all the algorithms and all the signals and it knows more than you. And if you don't use that, we're going to take points off of you or whatever, the kind of backlash you would have gotten. So it's really great to hear that. Yeah. I think it's very important to, um,

have that option available for the end users, right? So that, you know, you have a lot of people in your workforce that are experts at what they do and they've been doing it for years and years. So the tools that we're introducing are there to aid our workforce and our employees. So that's something that we've always kept front of mind as we drive our AI agenda at DHL. So you mentioned the users. How much do users need to understand that this is AI versus just a computer program? Yeah.

How do you get people to accept some sort of recommendations? Do you do a lot of training or how does that work? Do they need to know if it's AI? I guess is one way of phrasing that. I think people are interested to know if it's AI, but a lot of the time people just want to, you know, get on with it, right? So our customers, you know, they ask us for a solution. They don't want to understand in detail how that model was developed. How did we develop that algorithm? What type of techniques were used? They just want something that works, that's reliable, that

you know, is at the price point that meets their needs. And the same goes for our operations, our kind of end users of some of our AI tools. They just want to have something that's easy to use, that makes their lives easier and that they can trust. And if all those things are there, they don't want to, you know, dig into understanding what latest ML technique was deployed to make that happen.

A key part of the success is the change management. Many of the technologies that we're introducing into our operations, it's designed to make the lives of our workforce easier.

So I think in the past, you know, change management, yes, it's important. Yes, we need to focus on culture, changing communications, changing our processes. But over time, I think we've learned as a company just how important and how critical change management is, especially when you're introducing cutting edge AI, cutting edge robotics. It's completely new forms of human machine interaction collaboration. So that is something that is always top of mind in our innovation initiatives.

Could you share some other examples of use cases with AI? Yeah, so I have a couple of ones that are really exciting, actually, that ties in a bit to keeping humans in the loop. So we're also working with a startup on implementing AI-driven route optimization in last mile delivery and pickup. So there, again, we look at leveraging the data, looking into the route, looking into other external factors to optimize the best path.

for pickup and delivery and these pickup and delivery requests come throughout the day. So it's constantly optimizing the route. And our drivers who then actually get the recommendation on their kind of tablet or on their phones in the vehicle, they can either choose to follow that recommendation or they can actually choose to not follow the recommendation because they've driven these routes for years and years and some of them will just know the best way.

for various different reasons. So they're able to follow it, not follow it. If they don't, we can then try to understand why and again improve that algorithm for maybe a driver that's brand new and doesn't have that tribal knowledge. That's really a great example.

Somebody was making that routing before now, and then you've introduced an AI element to it. And I'm guessing a lot of that may have been automated before. But how do people react when you say, all right, I want you to follow what this computer is telling you to do? Are people thrilled? Are they angry? What's the reaction with people?

Always like to say you cannot trivialize, you know the people aspect I mean the AI can make a recommendation but it's actually people that are going to take the action, right? That palettes not going to just move by itself somewhere now that it's come up with this recommendation So with a lot of these projects that our team do we try to make sure that we have the right people at the table so it's not just the innovation team and

the leadership, but it's also people on the shop floor that'll actually be using the AI as an end user. So we try to get their buy-in very early on. And then we also give that option to say, actually, you know, the recommendation is incorrect, or I think this is the better way of doing it. So we allow that option so that we're not forcing everybody to follow that recommendation, but we still give the freedom to, you know, have people make their own choices as well.

Gina, I want to build on that comment around, you know, your customers want it to work. They don't want to necessarily understand all the details and then tie it to the teams you have at your innovation hubs. What are the kind of attributes or sort of personality types that you're finding your technical folks must have to be able to thrive in this kind of an environment?

I always say for our innovation managers, they're the ones that go into our operations and work with customers and partners to bring these projects to life. There are three kind of success factors. One is that they are able to have a deep understanding of technology. I'm not a technical person myself, so I always say that as a disclaimer, but I really make an effort to keep up with the pace of technology and try to understand different concepts and learn that pretty quickly.

The second part is understanding our operations. So one of the big kind of no-nos of corporate innovation is just sitting at an innovation center and being disconnected with the realities of your business. So we really make sure that we're out there at the operations learning about some of the challenges, talking to our people on the front line. And the third one is being close to our customers.

So being able to communicate some of these complex ideas and concepts in a way that our customers can digest them and translate that into business value drivers that our customers will embrace and want us to embark on and work together with them on. So it's kind of, I would say, a mix of an individual that has a good technical background but can really clearly communicate these concepts, get buy-in, and also down to earth that they can work in our operations on some of these projects. Yeah.

Just a quick follow-up. Who initiates these sorts of projects? Is your innovation team looking for them, scouring, trying to like, you know, we've got some cool tools. Where can we use them? Or do you have people approaching you saying we've got a problem? Which directions are these flowing?

I would say it's pretty organic at DHL. So sometimes it might start with the use case, right? So it might be a business unit saying, hey, we have this specific challenge. Like, what are some solutions to solve this? Other times we work very deeply with a whole host of different startups and there's a new one that comes by and we know that it makes things more efficient. Then we can find a problem to solve there. So it comes in different directions. Sometimes it's our team. Sometimes it's our business unit. Sometimes it's our customers.

And sometimes it's just a partner that's come up with a really groundbreaking solution that we know holds a lot of potential in our business.

As a professor, we call that the E all of the above answer. Coming from everywhere. So what's exciting about this to you? What's fun? What makes you dread getting up and going into a project? What makes you excited about going into a project? What's fun about or exciting, if anything? The exciting part of AI, robotics, and some of these other topics that I work on is

It is truly shaping the future of logistics. So, you know, some of the first robotics projects we did back in 2016, it was one of the first handfuls of robots we were putting into our warehouses. And then four years later, it's, you know, one of the highest topics on the agenda of our business units. It's all about, you know, how can we leverage new automation in our warehouses?

So I think that's always exciting that some of these early proof of concepts and pilots we do, they might be the first for the industry. And then several years later, they become the norm and it's just a way of doing business. So that always keeps it really exciting. And then, of course, working with some brilliant individuals, both within the company, but also with our partners, that always makes life exciting day to day.

I feel like some of that's the curse of AI is because it's all shiny and new, and then suddenly it's just what everybody's supposed to be doing and it's normal and you always have to be searching out for that next cool thing.

I think if we went back to the 17th century and showed someone spellcheck, they'd think, oh man, my quill will actually check, you know, underline with red when I misspell a word. I mean, that would be sorcery. But now, you know, if the paper doesn't practically write itself, we're bored with it and it doesn't really seem like cool technology. Is there any cool technology coming that you're fired up about or you think that you can apply to do something? You know, what's short term on the horizon that's exciting? When it comes to AI, I think the...

Evolution of some of our analytics services evolving into more advanced AI will be really exciting. So to give you one example, we developed some years ago a supply chain risk management tool called Resilience 360, which is very timely now because of everything that's happened this year. So with Resilience 360, it's a tool that alerts you if there's a risk that's going to disrupt your supply chain.

And it's a true kind of big data analytics lighthouse tool at DHL. But we never could really predict the risk and then kind of quantify what that risk will do to your supply chain. And there we're now working on leveraging AI and taking that to the next level. So I think that's a really exciting space. Thank you for taking the time to talk with us today, Gina. This was really great. Thanks, Sam. Thanks, Shervin. It was great to talk to you.

We really enjoyed talking with Gina. Shervin, let's recap a minute and talk about what we learned. One point is about AI being a big change. It's not just a tech thing. It's about change management, and she emphasized that.

Yeah, I like that point. And I'm going to borrow that quote from her that the first day for AI is the worst day because it very much talks to how difficult it can be. And so, yes, change management is critical, but it's also going to be difficult. And she talked about the process that she follows or she's created that brings the users and the operators into the design phase early on so that

They're not surprised by what AI ends up creating down the line, but they're very integral to the creation of it. They evolve it. They have the right expectations that it's not going to be perfect. We're working with this. We deploy it. We test it. We see how it goes. And so I thought that was really, really elegant how she talked about it's got to be a learning process and it's got to be

with the right expectation setting, but more importantly, the users have to be involved on an ongoing basis, not just at the end when the technical folks have built something and they're forcing it down to the users. I think it requires patience. There's a learning process involved, and if the organization is expecting things to run well the first day, then there's going to be a lot of disappointment.

Yeah, and I love that point because without patience and without that inner active ongoing engagement of user, the innovation center and the evolution of AI, there can't be a transformation. But if that process is being done cohesively, then...

The sky's the limit. That's when the innovation center can begin to actually completely change processes and create new ones and really change the way people work. Setting that expectation that it will get better from that worst day. Maybe that's a good benchmark to start with. We hope that it doesn't get worse. It's going to get better after that first day.

But the other part of that is the idea that AI itself can change the organization. AI can be a force for change. Jean was in the, she's in the innovation group and their charge is not to put AI across the organization. Their charge is to innovate.

And it sounds like AI has been really instrumental in changing the culture around innovation at DHL. Yeah, no, I totally picked up on that as well on several dimensions. One is that the users of AI are, you know, you asked them 10 years ago, how was it being done? Well, it was being done manually. And how were people going down those routes? Well, they were all deciding based on their own judgment. Of

Of course, today they have the AI telling them something. It's not being forced on them. But they have the benefit of that signal and that recommendation. And if they think they could do better...

they will know whether they did better or not. And if they did better, then AI will know that it could do better. And that process itself is introducing this change you're talking about, Sam. So I thought that's really an interesting way that they've set that up and they are scaling it across different use cases. She used the word recommendation a lot, and that was nice. She didn't talk about the solution that the system offers. She talked about recommendation, which is a very...

collaborative working together approach. I thought it was a great conversation. We're looking forward to our next episode with Matthias Ulbrich from Porsche. Please take the time to join us. Thanks for listening to Me, Myself, and AI. If you're enjoying the show, take a minute to write us a review. If you send us a screenshot, we'll send you a collection of MIT SMR's best articles on artificial intelligence, free for a limited time.

send your review screenshot to [email protected].