Data & Digital Twin Drive Real-Time Predictive Analytics

MICHAEL KRIGSMAN: We’re talking with Satyan Parameswaran, the President of Information Technology at UPS. He’s sharing with us an inside view of the way UPS uses data to manage their enormous network, , shipping millions of packages every single day. 

Satyan, welcome. How are you today? 

SATYAN PARAMESWARAN: Michael, I’m doing great. Thanks for having me. 

MICHAEL KRIGSMAN: It’s great to see you again. Satyan, we all know what UPS does, but give us a sense of the scale at which UPS operates. 

SATYAN PARAMESWARAN: UPS, we operate on a really large global scale. What I mean by that is we operate in 220 countries and territories, either we are there directly or through our partners. And that’s a lot. We, on a normal day, deliver 46 million packages and documents. If you just multiply that, last year alone we handled and delivered 6 billion parcels and documents. And the technology and the workforce and the logistics of that piece are needed to make sure that every one of them is handled as if that’s the only package or parcel to put a smile on our customers is our pride and our responsibility. 

MICHAEL KRIGSMAN: It’s just an enormous operation, and you’re managing, dealing with so much data. It’s an enormous scale. How does data support your operations? 

SATYAN PARAMESWARAN: So the fundamental principle of a logistics is making sure right things, the right resources, right effort happens and brought together at the right time. So it’s a very nice complex but efficiently choreographed dance. For that, we need the technology. And the technology is not just there to bring things together. You need to know what’s going to happen at what time, what resources are available at what time, and how many of those resources are available at what time when things are going to happen. 

So collecting data about the events and making sure that you bring them together and provide a relevant context to that data and then making use of it is the act of using data for running the logistics network. And that’s where our capabilities of using a modern state-of-the-art data analytics-driven methods and platforms is of tremendous help to us. 

MICHAEL KRIGSMAN: What kind of data do you collect? 

SATYAN PARAMESWARAN: We love data. If you are in logistics business, we just love data of any kind. We love data about the customers. Like, Michael, how many packages he has shipped in the past? What kind of packages are shipped? Where did he ship? And what kind of packages Michael has received? Who is sending those packages to Michael? And does Michael have any preferences in receiving those packages? Does he always want them to be left at the front door or at his neighbor or leave it at an access point so that he could pick it up on his way back home? That’s on the customer side all about the customers. 

Then coming into the UPS network when the package comes in, when did it come in? Where did it come in? How heavy it is? How valuable it is? Like, because it occupies the shelf space. And where it is supposed to go? When it is supposed to go and what time? So all about customers’ preferences and the packages that they trusted UPS to handle. We collect that data. 

And since I said that we handle the 6 billion packages and documents last year, each and every package and document would have a digital signal. And the customers who received them also provided the so intricate signals to UPS on when they are going to ship, when they are going to receive, and what they want. And all that data is used for managing the network. That’s the kind of data I’m talking about. And we have been doing this for 117 years. Of course, 30, 40 years back, the things were done very differently, but we collect data and we make use of data to make sure that we do not disappoint our customers. 

MICHAEL KRIGSMAN: So it’s a combination of real-time or near real-time data together with static data, such as customer preferences, for example. 

SATYAN PARAMESWARAN: Absolutely. A near real-time, pseudo real-time, and the passive time. And it’s all temporal in nature because the package when it goes through the network, after it gets inducted, many times it may stay there for a day, sometimes for a couple of days, depending on the service. Sometimes, you might print a shipping label. But you might give the package to us maybe tomorrow or day after tomorrow or whenever you have time. 

So the different data has a different context to us. And for logistics, time is of the essence. So having the data and having your relevant context with respect to the time sensitive nature of the data is of extreme value to us. 

MICHAEL KRIGSMAN: You built a set of tools and a platform to handle, to collect, to manage this data. Would you tell us about that? 

SATYAN PARAMESWARAN: I would be delighted to. It’s called HEAT. We love acronyms. Like every other large corporation, we love our acronyms. HEAT stands for a Harmonized Enterprise Analytical Tool. And what it does is it collects billion data points every single day from the customer, who provided, hey, this is what I want to ship, from the customers, who say, this is where I want to receive. And then the packages go through our network, the scans, the sortation equipment, industrial automation equipment, our package cards, our service providers, our trailers. Every one of them, including our aircraft, the containers we move, every one of them emit digital signals. We collect that data and process through our HEAT platform. 

HEAT platform is a representation of digital twin of our physical network. Why do we have to have a digital network? To make decisions and to know what’s happening where. Yeah, you can always run few reports, and the report might give you a partial picture. But having a living, breathing digital twin will help you to exactly know what’s happening to the network. 

Also, much more importantly, it’s the single source of the truth. When you have the digital twin, this is not like a four or five people are looking at the elephants from different angles and say, hey elephant looks like a tail, elephant looks like a pillar. No, this is a single source of the truth, where all the decisions could be made of it relevant and real time data. So HEAT is the platform, your proprietary advanced analytics and machine learning-based mechanism we built to handle all the data that is coming from our operations to be managed and processed so that we can have a digital twin. 

MICHAEL KRIGSMAN: So HEAT is the digital twin? 

SATYAN PARAMESWARAN: HEAT is the digital– heat is the platform that is the digital twin that helps us to make decisions. Yes. 

MICHAEL KRIGSMAN: I always think of a digital twin as representing a physical object, a digital twin of an airplane or an airplane engine, for example. In this case, your digital twin is representing the dynamics of packages moving through your system through your network. So that’s quite fascinating. 

SATYAN PARAMESWARAN: Yes, this actually represents the physical world. What I mean by that is we have 2,000 facilities across the globe, give or take, like, 2,010 2,020. Those facilities, they all process packages. They receive packages. They sort packages. Some of them are intermodal hubs. Some of them are destination package centers. And they all have a temporal attribute. Are they doing a late evening SOC? Are they doing your morning pillow? Are they doing an intraday SOC? 

So they all have different sets of contextual meaning to their operations. They are all play back the data. So if you step back and look at it, this digital twin will show you, these are all the 2,000 plus facilities. And this is what it is doing now. This is what it is expected to do like maybe eight hours later or 24 hours later. And based on the inflow of packages and the inflow of the expectations, this is how it’s going to behave in the future. So it is like a living, breathing representation of our physical network. Does it make sense, Michael? 

MICHAEL KRIGSMAN: Yes, it does. Then I assume you’re doing predictive analytics over the next period of time, whether it’s minutes, hours, potentially even days in some cases. 

SATYAN PARAMESWARAN: Yes, we do. HEAT platform does one thing very well. It provides the ability to forecast what is going to happen in a given node. Meaning, hey, this hub, be prepared to handle 47,000 packages on– say, today is Monday– Wednesday evening. So that the center manager, he or she can plan how many resources will I need to start those packages. And these kind of projections keep evolving as we get closer and closer. 

Today, I might say, you may do somewhere between 45,000 to 47,000. Tomorrow as we get closer and closer as physical things that materialize, we may come and refine that to a very high accuracy. So we help the individual operators to run their network and plan their network in a much, much better fashion. Going back to the point, we have to deploy resources at the right time so that the network can be managed properly. HEAT is the one that helps us to do that. 

MICHAEL KRIGSMAN: How do you use that data? Meaning, how do you transmit the results or give the results or the guidance to the operators in your various facilities around the world? 

SATYAN PARAMESWARAN: Well, that’s a great question, Michael. That is a great question because many times people may have a misconception that the digital twin lives in a very tall building, and then it is doing all the magic by itself. No. Here at UPS, we use technology to marry the digital and physical world. 

To have a digital twin to be of a meaningful and useful help to the operators, you have to go back and look at what do these operators do every day because these tools should be embedded into their work process. So if you are a preload dispatch supervisor, you come and you are given the forecast so that you can plan. That’s how it is used. It is not sitting behind the scene, hey, it’s going to happen. So we had to go and modify our operational processes to make sure that the digital twin is seriously assisting them to run the network. 

MICHAEL KRIGSMAN: In what form do they get this information? Do you give them a PDF? Do they have their own dashboards? Do you send them a report? How does it actually work in that sense? 

SATYAN PARAMESWARAN: So this is going to be a classical IT guys answer– all of the above. We do have very eye-catching dashboards, starting with the map of the US. And then there, you can go drill down, where are the hotspots, and then go look at it. And some of the work dispatchers need certain reports on the data dumps to be provided so that you can dispatch. So depending on how the work is enabled, we provide multiple ways for the digital twin to interact with the operators. 

So we have dashboards. We have visual representations. We have reports. We have automatic data feeds. All of the above is the answer. 

MICHAEL KRIGSMAN: So you try to give the data to the recipients, which are people, inside operators, inside the UPS network and whatever form will make it easier for them to consume and make practical use of that material. 

SATYAN PARAMESWARAN: So let me give you a different kind of answer because I do not want to undermine what we do. The digital twin just won’t give data so that it could be left up to the interpretation of the individual. We don’t take that chance. Because human brain is so precious, so it can make a lot of decisions on the fly. But it is also driven and influenced by emotion. 

So the HEAT platform, in addition to giving them high quality data, it also provides a prescriptive action that could be taken. So it goes. And there are occasions where we actually can tell, hey, there will be a 17 to 20 early AM packages are coming your way tomorrow morning because these are the packages that needed to be delivered the well before the normal business day starts. So staffing that form is an utmost of importance. 

So in those cases, the forecast HEAT platform offers is a directly plugged into the work planning system. In some other cases, it would say, you know what, you are going to dispatch a 400 drivers today. If you are dispatching 400 drivers, if you are off by 100 or 200 packages, it really doesn’t matter because they can be scattered all over the place. So we provide the data and the recommendations. 

So sometimes, it’s prescriptive. Sometimes, it is high quality data. Sometimes, it will act on it directly. Does make sense? 

MICHAEL KRIGSMAN: Yes. And I’m glad you explain that. So it’s not just static data reports, but it’s actually recommended actions to take interpreting that data. 

SATYAN PARAMESWARAN: Absolutely right. Absolutely right. And to gain confidence that these recommendations are valid, you have to prove that it is good. And the accuracy we are talking about is not in the low ranges of 70% or 80%, no. We are giving like a mid-90s and high 90 percentage accurate to forecast. But as usual, anytime you start, it will be difficult for them to, wow, is this really right. But over the period of time, we have made sure that these are good. And we just take care of it. 

MICHAEL KRIGSMAN: That’s a hugely high level of accuracy. 

SATYAN PARAMESWARAN: It is a very high level of accuracy, but we are always a constructively dissatisfied. See at a macro level, yeah, it can predict the center that services the packages you receive in Boston. Say, it handles the 8,000 packages. At the macro level, it might look mid-90s, higher 90s might be good enough. But when you break it down, there might be [INAUDIBLE]. There might be a section or territory where it might be a kind of low in volume and some other territories might be, all of a sudden, how to handle like a surge volume. 

So we are always on the lookout to go drill at the macro and the micro level accuracy improvements. It is a journey. It will be a never ending journey. 

MICHAEL KRIGSMAN: Yes. It makes sense. What’s the volume of data that you’re working with? Must be enormous. 

SATYAN PARAMESWARAN: So let me give some simple math in a very different way. On a normal day we handle 26 million deliveries. These packages gone through our network, a small portion of them for one day, some of them are for two days, some of them are three days. So on any given moment, the network keeps the processing the packages. 

So we typically say, we might handle like 50, 60 million packages in the network every day because some just got in, some are leaving, some are in transit. And with all the industrial automation, and all the customer expectations, and all the directives because you may just realize that you are not going to be at home tomorrow, you may put in a request, hey, hold the package, I will call, or redirect the package. So we receive all those things. 

So we are handling close to a billion events per day. So billion events help us to manage the network better. Now, let me take one step further like, why do we do this? We are a service industry. We are known for only one thing– delivering packages and other. Reliability is always the one that speaks for us. 

With our new motto, we need not be bigger, but we have to be better. Because we are going through the investment cycles. And right now, our CEO, Carol Tome, has made it very clear that we have to be better, better at providing the service, better reliability, and most importantly, how to contain the cost. Because we are a company, we have to make profit to survive. 

So being better means, without compromising the quality of the service, how do we improve the shareholder value? So this is where HEAT platform comes. It helps us to manage the network, which is the living, breathing thing that helps us to move the packages to be delivered so that we do not disappoint the customers and do not disappoint our investors. This is where HEAT comes into picture. 

MICHAEL KRIGSMAN: Satyan, you’ve kind of alluded to this, but can you be a little more explicit in terms of how does the HEAT platform allow you to provide a better customer experience while at the same time increasing efficiency, which is to say reducing time and reducing cost? 

SATYAN PARAMESWARAN: OK. That’s a great question. So like I mentioned, when you have 2,000 facilities across the globe, and we run the integrated network. In some senses, it’s like a hub and spoke system. And if you visualize, some people might look at it like a fishnet. 

You need to make sure that every node, every touchpoint is a running fine. Otherwise, the network will lose its integrity and efficiency. So the art of HEAT platform is making sure that we can react much, much better and quicker so that the network keeps running. 

I’ll give an example. Say you have a weather event. Texas had a really unseasonal high volume of winter storms this November. Several infrastructure capabilities were compromised. We couldn’t move. But if our network has packages flowing through parts of Texas, we cannot tell our customers that, hey, if your package is going from Atlanta to, say, San Francisco, but because there is a weather incident in Texas, we are going to be affected. That’s unacceptable. 

So HEAT platform, when it looks at how the packages flow through, it helps us to react quickly so that we can avoid the compromised spots. Because we live in the physical world, there are things that will affect the physical world, whether it is weather, whether it is floods, or even somebody digging up a hole can cut the fiber line. The communications could be cut. There could be power outages. 

So a lot of things affect us. HEAT helps us to react better. The moment we have a better handle on the way network can be managed, it drives customer expectations to a different level, because they know UPS has the skill and the scale to move the packages through the network, coping with all the challenges that can be thrown at us. 

MICHAEL KRIGSMAN: And the HEAT platform then incorporates potential issues such as weather, or supply chain bottlenecks, or airplane, or shipping disruptions. 

SATYAN PARAMESWARAN: HEAT is a managing on a day to day basis the internal aspects and the elements. Some of the things you mentioned, like the supply chain disruptions, they are all the input. So if we anticipate, oh, we are expecting 8,000 packages at a junction point. If it doesn’t happen, we will know, and we will react to that. 

Because we would have had your forecast for how much incoming volume is expected at that node. When it doesn’t happen, we have the ability to react. The airplanes are part of our network. So they are already embedded. We exactly know when the flights are supposed to take off, when they are going to be unloaded, when they are going to be reloaded. That is part of our network. So HEAT is the very integral part of how we manage our network. 

MICHAEL KRIGSMAN: I see. So all of these potential disruptions are inputs into the HEAT system, and then presumably, you’ve got such a large body of data that you can look back historically and then do predictive analytics based on the combination of historical data and the real time data about the situation as it is right now. 

SATYAN PARAMESWARAN: Absolutely. I will not shy away from expressing the complexity. One of the most a complex thing to do is pick up volume forecast, because we are dealing with the customer behavior. It could be influenced by so many things. So our philosophy is, yes, we have the historical data, we have seen this type of behavior, and we are expecting some volume to come to us. 

But the beauty of managing the network is how quickly we can recognize that whatever we expected is not actually happening, and how do we cope with that so. When you bring these two things together, that’s where you get the smart digital network. That’s why HEAT is coming very, very handily to UPS. 

MICHAEL KRIGSMAN: I have to ask, what are some of the technology challenges that come into play when you try to build a platform like this? 

SATYAN PARAMESWARAN: Couple of things. We are dealing with lots of data, so whatever platform you are going to build, it needs to have the capability to absorb, process lots of data, and scale up quickly and scale down. So that’s one part, a very highly elastic scalable platform. Because like any other project, you don’t start building your project like a HEAT on a really large scale on day one. You start small. You take a couple of centers, and then you create models, and then you slowly elaborate. 

So the scalable platform is a number one priority. So for that, we went with Cloud. So HEAT is hosted in Cloud. The second significant part of it, which I would say anybody venturing into building these kind of solutions should be extremely careful and cautious, is data quality. Make no mistake, tons and tons of data does not mean that you have high quality data. The quality of the data is a very, very contextual thing. 

You may think that you have high quality data, but if you are trying to get to the mid-90s, higher 90s level of accuracy, the data quality and the meaning of the data is extremely important. When you have a systems that are built over several decades, and then you are running a large network, different data that comes to you will definitely have different quality. 

If you merge every one of them together, the act of making sure that the data is relevant, and that you can make some interpretable and the meaning out of that is a huge challenge. Some of the earlier things we went through are understanding that data quality differences and temporal nature across several sources. 

MICHAEL KRIGSMAN: Satyan, we’ve been talking about the technology and the types and the composition of your data. Can we shift gears slightly and discuss the nature of the team that’s required to build this kind of platform. So can you tell us about the types of roles that you have and the composition of your internal team? 

SATYAN PARAMESWARAN: Michael, that’s a great question. Because many times when people ask me about how did you guys build HEAT, many of them expect, oh, we have a high quality technology team that does everything. That will be a great mistake if the team is made up of only the technologist. 

Since we are trying to create a digital twin, which has to mirror and reflect the reality, the process folks, who are the process engineers, who helped to manage and define the network, are integral part of the team. When we started, we didn’t jump into technology head first. We started with the process, hey, let us to define what’s the process that helps the package to get accepted, inducted, sorted, scanned, transported to the network and then loaded and then delivered. 

What happens? How do we manage? So we mapped the process first. Then we are asked, if this is the process, what are the things that are actually emitting digital signals? At what point, and what’s the quality? If you bring them together, what would we see? Then go back and build the system. 

So our team is made up of process engineers, developers, and modelists, data modelists. So that combination is the one that made us and helped us tremendously to create a product that will actually work. Otherwise, it will look like a science project. If you don’t involve the process folks who actually know that physicality, it’ll be a science project. 

MICHAEL KRIGSMAN: So this ensures that there’s very, very close alignment between what the business needs to ensure that customer happiness, ensure that packages flow properly through the system, reduce costs to align all of that, those business goals with a digital twin so that they work together. 

SATYAN PARAMESWARAN: Absolutely. So process people, developers, the guys who can manage large amounts of data, and the data scientists who can create models. It is a team with the multi– it’s a multi-disciplinary team to create a twin that will help the business. 

MICHAEL KRIGSMAN: Are there particular challenges associated with real time data, since you’re working with so much of it? 

SATYAN PARAMESWARAN: Any time you want to handle near real time data, it is always challenging. The reason is that you might very fast lose the context of what the data is trying to tell you if you are not capturing it and interpreting it quickly. 

So say we get a signal saying this package expected to be sorted in this facility. And if not sorted in 30 minutes, it’s going to miss. If you have a process that takes like a 15 minutes to discover that, it’s of no use, right? So dealing with the real time data is always a race against time on what’s the context, how quickly we can process, and do we have a separate channels of process, so that they don’t stand behind in a queue with the lesser contextual messages. So it involves with looking at the network on what’s important. And how important your particular message is. 

MICHAEL KRIGSMAN: And does this focus on data– you’re so data-centric– inform the culture at UPS in any way? 

SATYAN PARAMESWARAN: Once again, you keep asking great questions, OK. If you ask me, we are an engineering company at heart. We are an engineering company that happens to have package cars and service providers so that we can deliver packages. What I mean by that is, if you are a small package logistics provider, you need to have that engineering mindset to break down the act of delivering the package and break it down and measure every single thing, so that you can go, improve, implement, measure, and track. 

So we are an engineering company at heart. So when you are an engineering company, every single method you develop will have your mechanism to count, this is what is supposed to happen, did this actually happen? If it didn’t happen, what should be the course of action? So we always get into that mindset. So when we start initiatives, when we go into what all the things we need to do to manage the network, the fundamental genetic behavior comes out, which is we are an engineering company. 

MICHAEL KRIGSMAN: It’s so fascinating because, of course, from the customer standpoint, you’re a package company. You know, we all know I’m on good terms with the UPS guy who delivers here where I live. And yet from an internal standpoint, you’re an engineering organization. It’s really fascinating. 

SATYAN PARAMESWARAN: Absolutely. That’s the beauty of simplifying things, right? We don’t have to disclose all the complex choreography we do to deliver the package. Our UPS service providers are the best face of UPS that interacts with our customers. You just can’t beat it. 

So they do what they do. Technology here at UPS exists for only one reason, to help them to do their job. So that we want to be invisible. We just want to hide all the complex things we do in the back, so that it looks very simple. Our friendly, smiling UPS driver shows up. He picks up the package and delivers the package. He just makes a couple of interactions with you, then move on. We just want to hide all the complex technology to the background. 

MICHAEL KRIGSMAN: Yeah. It’s amazing, it really is, the level of complexity that you’re dealing with. Satyan, as we finish up, what advice do you have for business leaders who want to work more effectively with the data inside their organization? 

SATYAN PARAMESWARAN: A couple of things. Any time you are starting a data driven initiative, rest be assured the it is a journey. You may think that this is what you want to accomplish. Yes, you start. Many times, after you go through, that might not be the most valuable thing you might discover. You should always be open minded to explore the ancillary parts and stay closer to reach the destination, or follow the journey. 

So that’s one. Be open minded. And there will always be lots of times where you would have gone through a very hopeful phase, and then it will come crashing down. That is a typical symptom of realizing the data quality you have. You might have thought, see, everybody lost their baby. My baby is the greatest baby. 

But the data quality might not be as good as what we all thought. So be prepared for contextual data quality working very closely with the business, the business mapping. Because you come with a tool, and if the tool is going to predict something 60%, it’s not useful to the organization. 

You should work with the business to make sure that how to improve the decision quality using your data science related projects, so that everybody can create a model. They all know that it’s going to work. You should never have that doubt in mind, oh, am I wrong 40%? That’s not the place you want to be. 

MICHAEL KRIGSMAN: Great advice, very practical advice. And actually, I need to ask you one final, final question, which is this. UPS was involved in this massive effort to deliver vaccines. What’s that been like? 

SATYAN PARAMESWARAN: OK, how much time do I have? Because I have a wonderful response for that. See, yes, pandemic hit the world, and everybody was racing against creating a vaccine. And once the vaccine was available, they needed a mechanism to deliver very reliably. 

While that race was on, we were always looking at how to make sure that the health care related packages can flow through the network with extreme visibility, so that we can control the flow and react if something happens with the flow. We were working on several smart label driven packages. Well, the label and the package itself will keep us self-declaring, hey, this is where I am. I just got sorted, I just got moved, without needing any human intervention. 

We were working on that, and then we were ready, and you know, this is a label, OK? This label was used and created by us when we were in the lab. Many of you might not have seen this label in this form, but then we evolved the product, and then we scaled it up, and we actually brought it to life, this is how it looks. 

This is the UPS premium gold label. Every one of them is your battery embedded RFID label. You put it on a package, it starts emitting signals. So we exactly know where the package is in our network, so that we can act on it. So the act of delivering vaccine was an effort that’s been going on from a different point of view for us. 

We didn’t anticipate that the world would go through a COVID related pandemic. We were just getting prepared for health care products and that merchandise is of utmost importance to a lot of people. How can we create a package with the extreme visibility? We were prepared. It just happened. We have the ability and technology to scale it up. 

So we deliver vaccines to more than 100 countries with a 99.99% reliability. Why? We were prepared, and we just had the right technology that was available to work at scale. So getting prepared and knowing how to run the network was of utmost advantage to UPS to do that feat. 

MICHAEL KRIGSMAN: When it comes to something like labels, do you build different kinds of prototypes? 

SATYAN PARAMESWARAN: Labels, if I look back, I think the first generation smart labels were created by UPS maybe in the early ’90s. It’s 6 inches by 4 inches with a MaxiCode. It’s even sometimes close to 90 to 150 bytes of data. So you just can it, the package will tell where it is going. 

Because from the network point of view, we are always pushing towards the target. So we started with that. And then there are multiple improvements on the smart labels, how much data it can carry, and now, we are dealing with the world of really, really smart labels, the RFID which can declare, and then it can carry much more data, and very precise instructions that can be buried into the label, so when you scan them, it can tell hey, make sure that you get the signature from Michael, so that’s where it’s going. So it keeps evolving with the nature of how much data can be carried on the label, and how it can self declare itself. 

MICHAEL KRIGSMAN: I love that self declaring labels. And on that note, a huge thank you to Satyan Parameswaran. Thank you for taking us behind the scenes at UPS. 

SATYAN PARAMESWARAN: Thank you, Michael. It was a great conversation, and good to be here. 

MICHAEL KRIGSMAN: Thank you. I really appreciate it. A huge shout out and thank you to Redis for making our conversation possible.