Leaders Shaping the Digital Landscape
May 19, 2023

The Amazing Utility of the Digital Twin

In an attempt to keep up with the AI frenzy, host  will hold an insightful conversation with , CEO and Co-founder of , this coming Thursday, about how AI is powering new value from digital twin technology. Bring tons of questions...

In an attempt to keep up with the AI frenzy, host Tullio Siragusa will hold an insightful conversation with Bob Rogers, CEO and Co-founder of Oii.ai, this coming Thursday, about how AI is powering new value from digital twin technology.

Bring tons of questions and chime in to the conversation. Save the date!

#ai #digitaltwin #technology #podcast #liveinterview

Transcript

Tullio Siragusa (00:12):

Good day everyone. This is Tulio Siragusa with Tech Leaders Unplugged. Let's get unplugged again. Today I'm talking with Bob Rogers, who's the CEO and co-founder of Oii.ai. Good to meet you, Bob. Bob, hey, with me today we're talking about the utility of digital twins and how AI is powering new value from digital twin technology. So, before we into this topic, I know for a lot of people, digital twins, are a new conversation and new phenomenon. Some others may be a little more familiar with it, but perhaps it'd be good for you to give us a little background on how you got here and why digital twins.

Bob Rogers (00:57):

Yeah. Wonderful. Well so my, my, I'll start from now and go backwards. So, as you said, I'm CEO of Oii.ai. And prior to Oii, I have been expert in residence for AI at the University of California San Francisco, where I led a team of data scientists who created the world's first FDA-cleared AI on an X-ray device. That's some very interesting technology that's been deployed by GE worldwide and is saving lives. So I'm very proud of it. I'm also a member of the board of Advisors of the Harvard Institute of Applied Computational Science. And prior to all of that, I was chief Data Scientist at Intel, where I led Intel's partner ecosystem for analytics and AI. So, worked with companies like Oracle, AP, Microsoft, IBM helping them build their technology to use intel, use intel processors, but also bringing in the needs and desires of these analytical and ai, ai companies into the optimization of the Intel hardware most effectively. So that was a kind of an exciting role. Prior to that, I was co-founder and chief scientist at Appixia, which is a healthcare AI company that was acquired by Santini in 2020. There was a little middle period where I ran a quantitative futures hedge fund for 12 years that I co-founded. I don't usually talk about that in this kind of context, but actually, we use digital twins extensively during that work. So there's something to be said there. And then the beginning of my career, I have a Ph.D. in physics, and for my Ph.D. in post and postdoc work, I did work on digital twins of supermassive black holes in other galaxies.

Tullio Siragusa (02:58):

Fascinating, all right, so we're speaking to a rocket scientist, a physic and also one of the luminaries of AI. So okay. So you got your early start of doing these digital twins for, you know, for space, basically, but your, your path took you down the AI path, so it's all coming together. I mean, you could have applied your expertise of AI towards anything. Why did you pick digital twins? What's unique about it and what's the opportunity as the market?

Bob Rogers (03:30):

So, I mean, the, you know, so just first to frame up, when I say digital twin, what I'm really talking about is a simulation of a complex system. So a computer simulation that can, can express the behavior of something very complex that we might need to understand what would happen if this thing happened or this thing happened. And so, way back in the early days, what I was looking at is when you look in space, you see objects that are in other galaxies that are just sparkling and crackling with x-rays and gamma rays. It's crazy. And you, you think, well, okay, I'm seeing something very specific. What could possibly be causing that? We had some theories that material was falling in, being pulled and squeezed right at the event horizon of the black hole, and then shooting out a tube of a tightly wound magnetic field. I simulated, those conditions, and I was able to actually infer very specific pieces of information about, the supermassive black holes that were creating these x-rays and ga rays that we observed. So the, the fine, the, the sort of the theme there is that whole x-ray, you know, that whole thing of material falling into a black hole is very, very complex. I can't just sit down and figure out exactly what's going to happen on the pencil and paper, but if I put a very sophisticated simulation of what would happen if this material fell in and this and that and the other thing I start to get a very nice view of what it would look like, and then I can match that to what I'm observing and I can infer something. So if you know to, to really answer your question, the beauty of the digital twin is that it helps you really get a handle on any complex system where you really can't predict everything upfront, what would happen if that became a very, very powerful tool as my career evolved. So when I ran my hedge fund, we did a lot of simulations of the kinds of risks and, and ways things could play out in the market to understand how, how, you know, where we needed to protect ourselves. We're making bets on how things are going to play out in the market, and it gets very complex again. And, if you don't have a way to simulate what are the different ways things could go wrong, you don't have a good risk management strategy. And so in the trading world, if you aren't managing risk, you're really not doing anything. And so, I just continued to see how important it is to be able to predict how something complex is going to behave. So if we fast forward to today with Oii.ai, we are an AI-powered digital twin solution to optimize the design of supply chains. So how does that, how does that fit into what I was talking about before? Well if a company is making a number of products in a factory and those products need to be distributed all over the world to many different customers with many different needs, then that becomes a set of connected networks. So there's product A is going for, from the factory to a distribution center. From there it's two more distribution centers. Each of those goes to three, and then eventually through logistics, it gets to the customer. Well, that's one part of the supply chain, but then there's another product that's using those same trade routes, those same warehouses, those same factories. So really you have a couple set of distribution networks that all connect together. You can't have too much stuff going into a distribution center because the warehouse will get full. You can't ask the factory to make too much stuff, cause you'll run out of capacity. You can't have too much stuff in inventory because you'll start to have discards. So this, this supply chain is this very, very complex beast once you put all the pieces together.

Tullio Siragusa (07:51):

All right. So let's unpack that a little bit. How is that managed today versus how it could be managed in a digital twin environment where you have the ability to simulate and optimize and essentially it's a safe space to play around and break things and, and figure out how to make it better?

Bob Rogers (08:09):

Exactly.

Tullio Siragusa (08:10):

You know, how do they manage that today and how big of a disruption is it moving in this other direction? What impact does it

Bob Rogers (08:16):

Yeah, it's big, yeah, it's huge, it's a huge impact. So what happens today is you'll have some supply chain management software that basically has some parameters set. So, the parameters are things like what's going to cause me to reorder? So, so imagine you're in a distribution center and you've been fulfilling orders from the distribution centers downstream from you. There is, there are some parameters that tell you when to order more products, how much inventory to hold, how much to order, and, you know, some of the other characteristics of how you're managing your particular view of the world. Now that's compounded over all the distribution centers and all the con connection links and all the different products. And so the way that's done today is literally you set some parameters by hand into the supply chain management system. And very importantly, one of the pieces of information is the lead times in your network. So, if I need to figure out how to get a product to a customer within, let's say, two months, I need to know that this leg typically takes a month. This lake here takes two weeks. And furthermore, I should know how much variability there is in that. Is it really a month or is it three weeks to six weeks? And what happens today is that the lead time gets put into a fixed system. It's actually typically the ERP system, which then feeds the information to the supply chain management software. But those parameters stay fixed forevermore. And so typically when we go into a customer, we've seen that they set those parameters three years ago, and both the lead times that they're using, but also the parameters that control how inventory should flow through the system to achieve their objectives are completely out of whack. A fixed system, basically.

Tullio Siragusa (10:21):

It doesn't take into account market changes, demand changes, supply issues, and all the things that come together. So exactly today, how's that optimized manually? I mean, it's kind of manual,

Bob Rogers (10:32):

Literally, literally. You put planters, put their finger in the air, and say, oh, I think we should have this much, this

Tullio Siragusa (10:39):

Much looks like this.

Bob Rogers (10:40):

Replenish on, this is the safety stack, this is how we should replenish. And then sometimes they put too much pressure on the factory and the factory says, Hey, hey, hey, you're ordering too frequently. You got to slow that down. And then they fix it, then they, it's just sort of sloshing back and forth. Through the entire factory distribution network. And so there really isn't a view to optimizing, the way these things are being run. And so that's the opportunity with the digital twin.

Tullio Siragusa (11:08):

Well, now, years ago, just in time supply chain was introduced, right? I mean, it was not necessarily set up with machine learning or AI, and I guess it's served its purpose, but times have changed. Things move a lot faster. What is the intrinsic value? I mean, it clearly it's the ability to simulate and do things, but can you also do predictive?

Bob Rogers (11:32):

Yes. And that's where the AI comes in.

Tullio Siragusa (11:34):

Okay. And then what about those predictions, how do you make them, does that trigger automatic changes or is there still a human component to validating it? How, how does that come?

Bob Rogers (11:44):

Fantastic question. So, so you'll see if, if you look online about things that I've, I've written or talked about in the past on ai, I often like to think of AI as augmented intelligence rather than artificial intelligence.

Tullio Siragusa (11:55):

I've heard that before.

Bob Rogers (11:59):

Yeah. Put tools in front of people, make, make the AI, do the parts that people are bad at, and then let the people do the parts they're good at. Sometimes there's a piece of knowledge that, you know, a human will know that the, that the system doesn't. But to go back to the sort of overall thing, so the predictive part, if you think about what a digital twin is good for, you can put any conditions into a digital twin. What if this happened? What if demand went up 10% next month? What if my lead time, from Shanghai went from 30 days to 90 days because of a backup at the port? Right? I can put those into the digital twin and see what the effect will be broadly. So what's the impact on my ability to serve my customers? Am I going to have products discarded, you know, expiring because they're sitting in a warehouse? Am I going to have a problem with my factory? The pandemic really highlighted the fact that these systems today are, are static, and in fact, just in time end up being this weird mix of quite often starving parts of the supply chain for inventory and having way too much in other parts. This is a universal thing that we've seen. So the idea is if you throw the right questions or scenarios at the digital twin, it will tell you what might happen. But then the question is, well, what scenarios should I throw? And that's where we use AI. So we're using AI to predict what's going to happen with the supplier the supplier performance, what's going to happen with the, with the lead times in the network and the lead time variabilities, which is actually more important. What's going to happen with demand? You've got to forecast, I guarantee you nobody's forecast is a hundred percent accurate, but to what extent is it accurate, and to what extent is it going to have major failures? We can actually predict that. So we're not predicting the exact demand, but we're predicting thousands of different ways demand is likely to play out using our AI. And then that gives you the ability to see, okay, for those thousands of scenarios, 90% of the time with these settings, we were fine. 10% of the time we had a catastrophic failure. Okay, that's not good enough. So now we're going to need to tune our parameters to come up with a strategy that's going to give us good economic performance, get the product to the customer consistently, but also not have any of these, any of these major failures. And so then you know, the nice thing is you don't have to do that tuning by hand. Our system can automate the optimization of the whole process. So you say, what's important to me is not putting too much pressure on the factory and reducing my inventory by 50 million. Well, the system will come up with a set of parameters to change from where you are today to what you need to be, you know, these, these specific parameters. And to achieve those objectives, it will tell you where the cost trade-offs are, in fact, even cost tradeoffs for carbon because, you know, you change your supply chain, it's going to change what kind of transport and what are the carbon outputs of that, of, of those transports. So it gives you those optimizations. And then going back to your question about humans, typically what we do is we let our customers go through and just validate the recommendations, and then they get passed on into the ERP system. So there's a human in the loop just to make sure. And of course, you can't automate it completely. I think there's a time, what I've seen with other automation projects is there will be a point where people say, oh, wow, for this 80% of the cases, I want you to just do it automatically. I trust the system, it works. And then there's these other 15 or 20% that I'd like to look at and put in my human 2 cents worth. And now you're sort of allocating the work problem.

Tullio Siragusa (16:02):

Can optimize those parameters. So I'm, I'm also curious, is it enhanced by external data too? I mean, you we're talking about the supply chain. I, I couldn't help but think, okay, how do you predict the weather and, and yeah, so disaster, right? So what are some…

Bob Rogers (16:17):

So what we do is predict, right now what we're doing is rather than predicting point events, we're predicting the probabilities of different size impacts and the shape, you know, what's the size and shape of, of different kinds of impacts. There's, there's absolutely every likelihood that as, as we evolve, we will continue to expand, the sources. In fact, we're looking at a partnership right now with a company that is literally tracking an entire data fabric of information sort of about supply chains and external to individual supply chains that may have a role. We've been working with a company that is part of a very large container shipping company. You know, one of the, you know, you see the big ships go by with the big names on the sides of the containers. They, there are, there are very specific things that can happen that will impact the ability to load containers on and off of ships and ports and how, if one gets backed up, where does the traffic go? Those are things we can predict, not in the sense of next Thursday, suddenly it's going to go from 10 days to 80 days. But as something starts to change, we can predict what are all the, what are all the sort of collateral impacts and how would you put the whole story together. So we're using mostly the data that each organization is collecting about its own operations, but we have now worked with tens of thousands of different products in different markets and regions. And so we can kind of see how the whole story is evolving as a whole. And that gives us a very powerful platform for predicting what these impacts might be.

Tullio Siragusa (18:03):

Can it also help in terms of resource planning, identify, for example, over usage of certain warehouses or underutilization or routing, or based on the cost of fuel? I mean, there are so many elements, right?

Bob Rogers (18:18):

A hundred percent.

Tullio Siragusa (18:19):

So it takes into account all those elements to, but, but who sets the, is it automatically or, or you have to set the parameters for what you care about?

Bob Rogers (18:28):

So we, we get inputs at, at a very high level. So for instance, right now we're working with a company that's got a very specific inventory objective over the next six months. They need to reduce their inventory by you know, some number of tens of millions of euros. And, you know, that right now is being driven by the fact that interest rates are going up. And so the cost of that capital to maintain that inventory is very, very high right now. And it's, it's pushed itself out of, out of the, the, the operating range for the, for the organization. So they have this objective, and once we've got that, then we can, then the optimization can show them how to achieve that objective. Staying within things like, well, you can't, you can't break our service to our customers, right? Maybe in some markets, there's more room to, to more leeway, but in general, these are things that we, that we take in as data parameters up front. What is the sort of the rules of the game? Then we optimize and then they can do what we call what-if scenarios. So, suppose, they're thinking, well, we're going to do a promotion in the fall, and we need to kind of have an idea of whether we can actually fulfill, suppose the promotion is successful, we're going to get an uptick on these products in this timeframe. We don't know exactly what that uptick would be, but we have kind of an idea we can model that for them in a what-if scenario and then tell them, okay, if it plays out in these ways, this is how you're going to want to adjust your supply chain settings ahead of time. Because if you wait till it happens, you're out of luck, right? It takes weeks or months for products to move through some supply chains. So, you've got to, you've got to do that resource planning ahead for carbon. You can see, you know, what is the tradeoff if it can be for example, I'll give you an interesting example. It can be slower per leg of a journey to use rail to ship products. Hmm. Rather than say, let's say truck. However, there are trade-offs because real is also generally more reliable than a truck and it's got less, less carbon impact. So it, it, when you look at the trade-offs, it might be, well, okay, it's sort of a dead heat in terms of the overall performance for the supply chain, but hey, the carbon impact is way lower. This is getting me to my corporate objectives better, even though it takes longer. The irony is it, it's more reliable. So it actually is getting me the right levels of service for my customers and my cost is managed properly. So there's all these different trade-offs and all of those can be made visible. You mentioned factory or distribution center capacity. So if, if it, it can, we've seen this a few times where the ideal, you've got one distribution center that's really important to an efficient flow of material, and it's got a capacity limit. And so the best solution would actually be to have 10% more capacity in that warehouse. Well, we can present that to the customer and say, okay, you've got two choices. Here's, here's the solution that satisfies the current capacity constraint and it's great, but you could do better. What would it, you know, what would it take if you increased your warehouse capacity by 10%? And they can, they can then see whether it's a worthwhile trade-off warehouse capacity is often something that's somewhat elastic. So they can quite frequently say, oh great, well we do have flex space in that city. We'll add some more capacity and then they know what to do.

Tullio Siragusa (22:17):

It sounds like you are with, you're at the right time in the market with this product. Right? I would think in a bear market, this is probably the most important thing you could do. Yeah, yeah. To optimize the operation of your business, which directly links to cost, which ultimately translated to lower higher profits. But we're wrapping up cause we're up on time, but I have one more question that I think, could be interesting, you know years ago, <laugh>, I think it's maybe, maybe at least 30 years ago, peppers and Rogers started this whole movement about one to one you know, relationships with clients and personalization and customization, and we've seen tremendous growth in personalization engines, AI-related engines for that. But this seems like if applied correctly, you actually could get to an ability to do personalization and customization a lot easier because it, it's managed by machines that can optimize and figure out how to best support that. Are you seeing any thoughts about that? You know, cause sometimes you got technology that comes out, and all of a sudden you realize, oh, maybe that's possible. Now are, are, are conversations happening around that?

Bob Rogers (23:38):

It's a really interesting point and what we've, what we've, you're absolutely right that taking the tuning out of the hands of individuals who are going back once every six months or year or I mean, we've seen it as far as five years back, you know, that that's a painful process and you don't, the people who are doing it don't necessarily have a global view. When you want to optimize the system, you want to see the whole system. The people who are outputting the parameters tend to have a very local myopic view of the supply chain. So, so just from the point of view of managing that next level of complexity, this kind of technology absolutely makes sense. And actually, to your point about customization, we're talking to a large medical product supplier who has their product, each product line has many, many, many different little variations to adapt to each individual, you know, sort of prescription. And so they need to figure out how do we automatically and, and proactively route the right sets of those variations to all the different local distribution centers. Because traditional methods either result in way too much inventory that never gets used until it expires, or we're constantly direct shipping. And the problem with that is not only is it expensive, but it actually also causes a delay and causes us to have erosion in our, in our sales because if there's a delay, our competitor might get in there and take the business. So that is a kind of customization where each local segment of the market really needs its own tuned inventory policy and you just can't do that with humans.

Tullio Siragusa (25:28):

Yeah, I mean it's, it's interesting you're speaking about pharmaceuticals in particular. I can see where this kind of technology could eventually enable manufacturers, to personalize even medicine not based on people's alley. Or what have you. There could be a nuance that's specific to that particular yeah. Patient. So it's, it's very exciting what can be done in terms of leveling up the service, managing resources better, and simulation optimization technology is really key. So yeah. Just stay with me as we go off the air. It's been great speaking with you. We wish you a lot of success.

Bob Rogers (26:06):

My pleasure.

Tullio Siragusa (26:07):

Sounds like anyone who has any European system or supply chain system in manufacturing should probably jump on this pretty quickly. Seems like a great opportunity. Stay with me as we go here in just a second. I want to announce what we got coming up next week. We're done for this week. We have three guests, which is exciting. And next thing we're going to have Tony Sumpster's the CEO o at Worksoft. And on Wednesday we're going to have Prashant Sarodi who's the SVP of automation at LPL Financial. Thursday we're going to have Amanda Blevins, who's VP and CTO of VMware, and Friday Piyush Malik, who's CTO of Veridic. So come join me next week, at 9:30 AM Pacific. We're going to have four great guests that we're going to dig in and learn some new things and see how these leaders are disrupting the technology space or basically leading and shaping the technology space. So it's been great to have you with me today. Thanks again for joining us. Have a great rest of your day.

 

Bob RogersProfile Photo

Bob Rogers

CEO & Co-founder

Bob Rogers, PhD, is CEO and Co-Founder of Oii.ai. Previously, he has been Expert in Residence for AI at the University of California, San Francisco, and a member of the Board of Advisors to the Harvard Institute for Applied Computational Science. Bob was also Chief Data Scientist for Analytics and AI at Intel, and was Co-Founder and Chief Scientist at Apixio, a healthcare AI company that was acquired by Centene in 2020.

Bob began his career with a PhD in Physics, developing digital twins of supermassive black holes in other galaxies. In 1993, seeing the future potential impact of AI, he expanded his research to include artificial neural networks, the progenitor of modern AI technology. He co-authored the book, “Artificial Neural Networks: Forecasting Time Series,” which led to a 12-year career as Co-Founder of a quantitative futures trading fund. He received his BA in physics at University of California, Berkeley, and his PhD in physics at Harvard. He has been featured in numerous publications including Forbes, Inc., and InformationWeek.