Using the Power of Data Analytics to Accelerate Digitalization Across the Supply Chain

In this episode, we discuss the front-runners of accelerating digitalization - from collaborative robots to software-based automation.​

Special Guests:

• Brian Gaunt, Senior Director of Accelerated Digitalization, DHL Supply Chain North America
• Steven Grover, Director of Data and Analytics, DHL Supply Chain

*Previously recorded at CSCMP Edge conference in Atlanta. 

Transcript

Speaker 1: 0:03

Welcome to all business, no boundaries, a collection of supply chain stories by DHL supply chain, the north American leader in contract logistics. I'm your host will Haywood. This is a place for in-depth discussions on the supply chain challenges keeping you up at night. We're breaking beyond the boundaries that are limiting your supply chain. Let's dive in today's episode is using the power of data analytics to accelerate digitalization across the supply chain. This podcast was originally recorded at CSC MP edgy 2021 conference. And our guests are Brian gaunt, senior director of accelerated digitalization for DHL supply chain, north America and Steven Grover, director of data and analytics also for DHL supply chain. So welcome to using the power of data analytics to accelerate digitalization across the supply chain sounds simple enough. Um, this is a hot topic as most of you know, uh, and one that we have, uh, two experts with us today, uh, to tell us a lot about, uh, what we're seeing in the market and, uh, what we're doing with our customers at DHL supply chain without further ado, I'll introduce Brian.

Speaker 2: 1:15

Good afternoon, everyone. My name is Brian gaunt. Uh, I am the senior director at DHL responsible for the coordination of, uh, accelerated digitalization. And I'll give you a brief snippet of what that is and my portion of the presentation. So it is looking at new innovations, both hardware and software oriented and coordinating those with the business. Uh, we are continuously innovating our company, um, continuing to strive to do business a different way. And this is part of our key strategy. And, uh, this is important. It's not just one team doing this. This is really our culture. And we've really progressed in this area in the last four years. You know, our goal is to be the leader and innovation sustainability in this space, uh, driving these demands for our customers and making supply chain even more digital slash automated as we can. Uh, this is core to what we do, and this is important because it really flows through all levels of our organization and all different business functions. So a big part of what we do at DHL is we focus on the things that are important to us. Uh, we drive this, um, process by focusing on specific use cases that drive value for our organization and customers and the process we follow for this. We have a very structured process around this. So we run this through what we call our innovation funnel process, where ideas start at the left-hand side and ideas can come from lots of different people, industry experts, um, our business and supply chain experts, um, vendors. And we drive these things through a very structured process. So we have dedicated teams in place that look at new processes. And, uh, we're not just looking at them because we think they're interesting. We're looking at them to solve real problems. And this is I think, one area that DHL does a strong job at, as you may have noticed of talking to other robotics vendors. Um, many times these roboticists don't really understand our business and don't understand supply chain. And, uh, we really co-developing collaborate to drive these solutions that are driving value for our space. And then ultimately after we innovate an idea and we put all the structure around it, we get to a stage called product position. So at this point in the product stage, we really understand the technology. We understand the production rates, we understand the value to the organization and we're really ready to scale this type of solution. And then that takes us to the deployment stage on the right hand side. So now, um, you know, we scale these types of solution to our 400 plus sites in north America, you know, where there is a use case and fit. So that's important part of the process, we take an idea, we put the right focus around it, and then we scale and we scale fast, uh, work fits. And there's lots of different tools, but that's, that's our process. Uh, we're definitely a multi-vendor type of company. Uh, there's lots of different ways to solve similar problems. So, uh, we have strong alliances with many different types of partners, um, that bring, you know, software solutions, robotic solutions to our business. And then this matching process that occurs in our product design or deployment stage, um, really goes to the opportunities identified. So it's more than just taking a look at these. We're making these real and rapidly deploying these types of solutions throughout our business. As you can see, you know, we've have over a thousand, um, go lives just in the last couple of years and these focus technologies. So I think that's the important thing. We're not just a research institution company, we're driving these solutions, driving them to real value in making them production solutions. And, uh, I'll give you a quick glimpse at how we organize these 12 categories, um, is obviously lots of things to focus on, but this is our kind of perspective on things that are driving value. So these 12 kind of categories here are the things that we feel is driving supply chain in north America, the most value system, picking goods to person wrapping robots, et cetera. I won't read everything here to you. Um, and I think what's important is these are more than just concepts. Uh, these are also driving to, to be live. And as you can see, intermingled in there, um, within our innovation process, it is a mix of hardware type of solution. So there is robotics, physical robotics, and then the software side of things, which is this intellectual, um, these algorithm and optimization stage and other things that are important from a software perspective. Um, focusing a little bit on the software side. I think this is the important side is as he began to talk to robotics companies, you'll at least I I'm coming across that most companies are now saying I'm not a, I'm not a hardware company. I'm a software company, even, um, Ford said, I'm not a car manufacturer, I'm a software company. So that same mindset has fallen over to our supply chain space as well, where the software side is really as important, if not more important than the physical hardware side. And this is just a sampling of some of the types of hard, um, software type of solutions that we're actively involved in, uh, machine learning, um, machine translation, robotic process automation, uh, natural process learning all these different software components come together. And I think the most important thing on this slide is the people in the associates. So you can go out and buy, you know, any one of these kinds of components, but having the right focus from your associate perspective and solving real problems within our supply chain is absolutely key. So the software components, um, obviously get their, their meat and dry from the ideas from our associates, and then they're driving value back to them to make their process more automated. So everyone's pretty familiar with when you think of the basic functions that a warehouse receiving storage, picking, packing, and delivery, um, what we've done is we've lined up all these technologies at the bottom to align to those different specific use case. This goes back to we're very much a use case focused innovation company. And what's important about this is all this stuff comes together. Um, with this bottom layer, we call it algorithmic optimization or resource orchestration. And this is important because we're tying together both the manual processes and the software and hardware processes and creating this feedback loop. Um, all of this information goes to our Dana Litt data analytics engine, and this is optimizing the processes.

Speaker 1: 7:57

Good. So I did have a question on this one for you, Brian, when you put this up in front of customers, what's a T you know, how do they typically react and then how does that discussion go?

Speaker 2: 8:06

Yeah, absolutely. So we show this a lot to our customers. I think, I think it goes over quite well. I think it's easy to resonate that you can see that we have a very focused process. Um, we're not just throwing ideas at different processes. It's a very structured impro approach to drive these key technologies to solving these key warehouse problems in w that's that's occurring. The other thing we're doing with these solutions is, uh, we're really not just automating a manual process. We're driving automation and complete, uh, creating an entirely new process because of the automation. And I think that mindset's really key.

Speaker 1: 8:41

Okay. And then how do you prioritize, I mean, you couldn't do all of this at once. I wouldn't think how do you pick your spots to, you know, start and

Speaker 2: 8:50

Go forward? I think that's, uh, that's the challenge. We do have some focus technologies. There's a lot on our plate and we have a relatively small team, um, considering, but we do focus on key solutions that are driving value. The biggest ones we focus on, uh, is, is items that are helping with our outbound pick process. It's we have the most labor, so solutions that help with each picking case picking pallet picking are our highest priority. Along with that, as our, as our software suite of things that Steve is going to touch on a little bit from a data analytics.

Speaker 1: 9:22

So I'm not going to make you pick your favorite child, but, um, if you could maybe pick two of these that you think have the most promise and sort of why, and if you have an example or two of where we've deployed them and kind of what you've seen from a results standpoint.

Speaker 2: 9:37

Yeah. I think, um, maybe I'll pick three, um, from a, a perspective we've had the greatest traction assisted picking, which, um, this is the locus type of solution has been, uh, very successfully deployed within DHL. We have well over a thousand bots and rising rapidly each month. Um, these really give us a ton of optimization around each picking the other area, um, that we're growing quite fast in is a space called indoor robotic transport. And this is the AMRs and AGVs around pallet moving. I think the vendors have finally got it right. At least certain vendors have finally gotten this right, that these are solutions that we are actively deploying in our warehouse. And then the third one, I think, would be the software side, um, going back to the importance of algorithmic optimization, those will be our three focuses.

Speaker 1: 10:28

Okay. And then, uh, maybe last question, and then we'll, we'll get Steven up, um, commercializing these things or the co or the cost of these kinds of technologies. Um, I know an example for, uh, locus that we covered on the podcast, um, uh, with Brooks running. Um, but can you talk about how customers are able to bring these things on at a, you know, reasonable cost with an ROI that makes sense, um, how that's evolved maybe over the last 12, 12, 24 months?

Speaker 2: 10:54

Yeah, absolutely. From a commercialization perspective, um, we've changed our approach. I think traditional kind of other way of older way of thinking would be look at that particular innovation and then try to commercialize it for that specific slight site. And that burden is on that specific warehouse or that specific customer. And we've taken a completely different approach. These automations now, especially mobile robotics are very scalable. So the funding for these are occurring centrally. And then, um, whether the vendor has a kind of a Raz model, a lease model, we do that internally. So the sites only pay a monthly usage charge. So we're not trying to commercialize the entire thing for that one site. And Raz stands for row, either resources as a service or robotics as a service. So it's a, it's a way you can get, uh, we, we take these kinds of fees and, and normalize them into just a straight monthly fee. Okay.

Speaker 1: 11:48

Okay, great. All right. Why don't you get off the hot seat and Steven? Fantastic. My name is Steve Grover.

Speaker 3: 11:53

I'm director of data and analytics for DHL. I've been with the company about a year and a half. I've been in analytics for about 30 years. What I'd like to do today is help you create a mental picture in your head about how you can think about machine learning and its impact to accelerate a digitalization. Think about what you need to be able to do that. One thing you need is you need intelligence. We're going to talk a little bit and intelligence, and unless you're going to have people constantly driving all these things that are digital, it'll probably have to be artificial intelligence. So we'll talk about that. We'll talk about automated machine learning because we need a tool in order to be able to do that learning. And we'll go through that a little bit. And then we'll talk about composable analytics, because if you have intelligence and you have a tool, you need data to feed it and you better have a mechanism to feed it really flexibly so that you're not constantly reinventing the wheel. And then if you have the technology and you have the data, then you better have a really strong process. So then we'll get into the machine learning life cycle and Alma lops, and what that's about. And then we'll talk about the machine learning value equation, because in the end, if the equation doesn't work right, if there's not value, the value is not greater than the cost. And what's the point. And we'll walk through three use cases, kind of give you a feel for what that looks like. So where does the intelligence comes from? If you're going to digitize your business, you need something to do the thinking for you. And that's where the artificial intelligence come in. Artificial intelligence has actually been around since about 1950. And it usually came in the form of some basic robots that were programmed with if then else type logic, very simple stuff to begin with. Then in 1980, we started a machine learning and that's where you created all the rhythms that took in data. And then based on that data, it would learn and figure out how to handle certain situations without being specifically programmed for it. And then in about 2010, we started with deep learning. And that was the ability to create, um, neural networks, which is kind of like the way your brain functions in order to create thinking machines. Today's conversation is about machine learning because that's really the strongest space right now, at least in transportation and warehousing, the tool set we've been using is we're using some auto machine learning tools. And the advantages of those are that think about who the people are. We can think of the business use cases. It's usually somebody in the business, usually a data analyst or a business analyst. They say, you know, if we could predict X, then we could generate value. Why? Because we know what's coming. So, you know, a lot of your data scientists, they also need to know programming. You have to able to do computer programming. You have to know the business. You have to know ML. You not have to know how to wrangle data. But if you're a business analyst who knows a little bit about ML and you know how to wrangle data and you know, the business, you use an auto ML tool. So that's real important for us. In fact, we are strongly embracing a citizen data scientists. If you've got this auto ML tool next, you need to be able to feed it some data. And depending on how you approach that problem, you can be reinventing the wheel, the data that feeds your, both the training, testing, and implementation of your machine learning projects. So what we focused on is creating a set of analytic assets. And then what we can do is take those analytic assets. So let's say we have a use case called warehouse safety. We can take those assets and we can organize them in a way and join them in a way that allows us to be able to create data sets for training and testing our models. Then when the employee gets ready to do the next use case, let's say this one's dock lock for those of you don't know, dock locks, a situation in a warehouse where there's no place to put a trailer against the warehouse, because they're all folds very bad situation to be in. Especially if you've got product in the yard, you need to fulfill orders. We can take those same analytic assets, put them together in a different way, enjoying them and create the training and testing sets in order to create those models. So we're able to use data and iterate very, very quickly. So you've got a tool you've got data. Next thing you need is you need a process to make it all work, right? So we start with the development and the most important part of the development phases, the ideation, and there's two key pieces. One is you need a problem. And then the based on being able to predict that problem, there's gotta be value. So to get past that first stage, you really have to have a good sense of what is the value. If we can solve that problem, then you wrangle the data. You go through your model building process, and to get out of this phase, you've got to evaluate that you're able to predict it well enough, and that you're able to prescribe a solution that will deliver that value. If you've done that, then you've developed a good machine learning model. At that point, you're ready to start talking about deploying the model. If you've got a good working model, you have to figure out how you're going to implement the solution, because there could be, um, training issues. You gotta make sure you coordinate with the customer. The customers have a big say in what you're allowed to do in their warehouses. Once you've implemented and you started go through UAT, you might need to tune the model. There might be organizational change issues that happen. Maybe you need a new role in the organization to be able to handle some of the issues or opportunities that come out of making that model. And then you're ready to go live and start using it in real life. And then once that's up and running, you better monitor it. Cause there's a bunch of different things that can happen. So usually you have a set of KPIs that you're monitoring, and if they go beyond threshold, it triggers some kind of activity to check and make sure that the model is still on track. Another thing you better do from time to time, make sure the users are still engaged with the way that's working and they haven't drifted into some old pattern. You could be driving efficiencies and people decide I'm going to do it the way I used to do it. And all of a sudden you've lost all those efficiencies, right? And the other thing you'd better look for is model drift. So model drift is when something has changed the cause the model to slowly come off course, um, and there could be a business rule model drift. There could be data drift. There could be ethical drift. Maybe you start out in doing some things, okay. And then maybe later on, it's not okay. And then, um, over time, you're gonna have to make some solutions adjustments. So this is kind of our three-stage lifecycle process. And currently we've got several dozen, um, machine learning use cases that are in various stages of this. I'm going to cover three of these with you today. Now I made a big deal out of the value equation, because if you think about it, if there's not a value, the sooner you can kill a project that doesn't have sufficient volume, the better off you are. So you got to make sure you've got to constantly look at the value and make sure it's greater than the cost. When you're looking at benefits, most people identify the first two easily, if there's a cost savings or there's revenue, but don't discount these last two risk reduction. If you can reduce your risk over time, you will get real value out of that. And the other one's Goodwill. If you do something for safety or something that helps out employees, the Goodwill you can create from a machine learning solution is really substantial. And then on the cost side, again, the first two people kind of think about development costs unless you're developing a model is costing hundreds of thousands of dollars. Um, you can usually not worry about the development costs. You're usually focused more on the run cost because once you're generating value, if your run costs lower over time, you're going to make enough money. So it doesn't really matter. But the two costs that look out for is that prescriptive activities cost. And for those of you are not familiar with that is, is, um, that's if, when you come for, uh, when you develop your solution, usually, okay, let's say I can predict that there's going to be an accident at a warehouse. Well, I have to do some things. I have to prescribed some activities to keep that from happening. And the other one is opportunity cost. If you're working on this machine learning use case, you're not working on this one, three use cases, predictive cycle counting in the warehouses, we do two kinds of cycle counting. We do a cycle counting where we go through all of the locations in the warehouse and identify what's in inventory at each one of those locations and then their cycle count or count back where when you do a pick from a particular location, they do, they do a count back to see what the remaining inventory is. This process identifies where the high error possibilities are and focuses on those locations. And what this allows us to do is they have much more accurate inventory, way less work effort and huge returns on this one second one is predictive warehouse safety. I was talking about that a little bit. All this is looking at all of the factors, training, tenure, uh, the Superbowl happening all look at all of those features in order to predict where those are going to happen. And then take actions either through a phone call or through maybe a talking to a robot to make sure that some activity does or doesn't happen to avoid an accident. And then the last one predictive absenteeism, this is really about making sure you can maintain staffing levels. We're not really worried about is, you know, Jane DOE or John DOE going to be there that day. They're looking at the, at the whole team and saying based on either, um, holidays, uh, religious events, whatever, we could see this kind of an add impact to our non-exempt staffing and then make adjustments in our staffing so that we allow for that whatever's going to happen so we can maintain performance and really, uh, reduce over time and improve our throughput. And it can really improve this as a Goodwill item, right? Employee stress. If you're on a team of seven and you see only five show up that day, you're like, oh, today's going to stink. I know that was fast. Sorry.

Speaker 1: 21:51

No good examples. I, um, I had a question around who, who do you talk to, to give you the ideas? So who provides the best input for, um, what could be a, an array of ideas? I would think

Speaker 3: 22:03

The best ideas don't come from data scientists that would be the first clue, the best ideas come from business people in your organization, people who are close to the front lines and see the activity. So the business analyst, the data analyst, that's why we're really big on the citizen data scientists, because, um, we're seeing our, some of our most interesting use cases like predictive absenteeism, like warehouse safety from those folks.

Speaker 1: 22:30

Um, and then back to that value equation bit, how do you measure that? And when do you know an idea is a bad one? Is there, you know, are there red lights that go off that you say it's time to move on? Or how do you, how do you sort of calibrate that?

Speaker 3: 22:42

Sure. So there's a couple of places where that typically occurs again. Um, you want to try to find, you want to kill an idea as quickly as possible. Not because you want to kill the idea because you don't want to waste your time because there's other ideas that could be of higher value. There's a couple of key pieces places. One is when you're building the model, if you can't predict well enough, you've either got to get more data. You don't have data that's predictive enough, or maybe it's not the right frequency or the right grain, but you got to make sure you have the right data to get a high predictive rate and other places and evaluate and prescribed. When you're thinking about those prescriptive actions. If the actions you have to take are so costly that they don't make sense, that's another place you can kill it. You can kill it in solution planning. When you're looking at the solution, you say, oh, in order to make this work, we really need to hire a bunch of people, usually buy a implementation and buy two and three in deploy. And for once you get past one, you're usually pretty, um, confident that it's going to happen.

Speaker 1: 23:44

Okay, good. And then maybe to bridge the two subjects together, um, you know, you talked about acquisition of data and I don't think we're lacking for that. Um, even pre automation when we're just sort of at the WMS level, you know, or TMS, we're sitting on huge amounts of information that are flowing through on a daily basis. Um, Brian, when you think

Speaker 2: 24:05

About more automating technologies, how does, how do you think about the data flow that comes out of those, you know, actually adding onto the pile of data. And then what do you think about from a data science standpoint of sifting through that down into something that's either usable or you've got technology that can, that can weave that into something of business value. We're definitely not short on data and probably you aren't either, but you know, our goal isn't to create a data swamp of things. So we definitely, when we look at data elements and data components, we want to make sure that we're pulling relevant data. That's going to drive us to the highest value. So the important things of feeding data out of these robotics solutions is to make sure we're really focusing on the key data and it's at the right level. Um, in some cases we, um, you know, actively summarize that up to a higher level. In other cases, we work at at the lower level, but I think it's important to really spend some time and not say, just give me everything you have and I'll work through it is to really focus that dataset down. So then we pass it over to Stephen's team. Um, they can be a bit more focused.

Speaker 3: 25:08

So yeah, I talked about that a little bit before a lot of folks try to, um, for example, you can take a data lake and put all your operational data in a data lake. And if you've got 500 sites across north America, you have, um, separate instances, the warehouse management systems, the labor systems, you've got multiple HR systems, you got multiple timekeeping systems and you tell your data Santos, oh yeah. Put together this use case. And they got to look across thousands of tables. What are the chances you're ever going to be successful? Pretty close to this. Yeah. Pretty close to this. But if you can create objects that a user intuitively understands what it is and make it something that they can use over and over again, you can, you can do things very quickly. Once you have this up, once people grasp how they can glue this together, because there's a big difference between I have to be a programmer and I have to do a join between two tables and SQL, most people can do a join, you know, join it's in the, the answers in the question, how do you join two tables? You use the term join, right? So that's pretty simple. Whereas programming and having to glue together, hundreds of, in some cases, thousands of tables, that's a much bigger task.

Speaker 1: 26:19

So you talked about the concept of citizen data analyst, um, because you know, these are fairly high flute and concepts, especially when you think about traditional warehousing and transportation. And I know they have impact, but I think, I think where do we find the people to do, uh, these kinds of tasks because they don't meet the, you know, historical profile of some of our folks. Um, where are we finding folks? I mean, I know you're relatively new. You've been around for awhile. Um, Brian, and when, you know, for younger people, what are fields of study or experiences that they ought to, um, consider getting so that they can kind of tap into this?

Speaker 3: 26:58

So, um, the term I use is citizen data scientist because, uh, like I said, I've been in analytics for 30 years. I've known lots of data scientists. A lot of them have PhDs, advanced degrees in statistics. Um, and they, and a lot of them have programming backgrounds. And most of them will be anywhere from 150$250,000 a year in salary. So there's usually very few of them you can afford to have on your team. And then you have the whole industry and business knowledge thing, right? So how do we grow ours? We have an analytics community of about 200 people in NorAm and those people all kind of know SQL. They all know the business because their number one job, isn't being a full-time data analyst. They've got to actually make sure the business is running. So what we do is we take and give them the AML tools. We give them this data. And so we're growing our citizen data scientists. That's what we're doing. We're not going out and hiring we're growing them because quite frankly, they're very hard to find in the market. And when you do find them, they cost a fortune and they usually don't have your industry expertise. So that's how we're fighting the challenge.

Speaker 2: 28:00

That's exactly it. I mean, we, we do hire PhDs and we do have those individuals on the team, but I think mixing the, um, ML machine learning knowledge with the business is absolutely critical. So, you know, we've found better success tapping into this broader 200 kind of pool of individuals and then bridging that gap and teaching them some of them more advanced analytics skills, um, in order to operate the tools, we do have both. Um, but I think the, uh, the ladder, as Steven said, is, has been more successful.

Speaker 1: 28:30

Yeah. And I have heard from our college recruiting, um, channels that the kids coming out of school now are more predisposed to this kind of, um, these types of skills and they like it. So, Brian, um, what percent of our customers do you think are today saying, I want a robot cause it's cool.

Speaker 2: 28:51

Uh, 100%.

Speaker 1: 28:54

And what percent of that? A hundred percent I can't can see the, uh, the, the, uh, the business value coming from.

Speaker 2: 29:01

I, I think, uh, I think our customers are the same, like anybody else on different spectrums from really understanding the technology. Um, some, some are driving it and really want to see it in their operation because the cool factor, um, we're able to drive and show the real Bennett business benefits of putting in those solutions as well and help bridge those gaps in conversation and drive to see why this, these types of solutions are really helping, uh, their business from a quality productivity, et cetera.

Speaker 1: 29:28

Okay, great. So, um, thank you gentlemen, for, um, uh, your presentations today. Really appreciate it. If you enjoyed the conversation today, please share it with a friend and rate us on apple podcasts. You can find us online at dhl.com/all business, no boundaries and follow us on LinkedIn and Twitter at, at DHL supply chain. We'll see you next time.