How a Robotic Dog Fetches Risk Data; Season 1 Finale Ask Me Anything

Where are advances in technology taking data and analytics and enhancing our ability to prevent potential losses? In this episode, host Pete Miller, CEO of The Institutes, talks with Nancy Greco, a Distinguished Engineer with IBM Research, about the evolution of risk data and artificial intelligence to predict and prevent losses that can disrupt operations. In the second half of the episode, Miller sits down with Matthew Kahn, President of Risk & Insurance, to review some of the key takeaways from the first season of the podcast and moderates an ask me anything (AMA) segment with Miller.

Segment 1: (02:3)  In the first segment, Pete and Nancy discuss her career and role at IBM, the drivers shaping the evolution of risk data, a shift in focus to loss prediction, using Boston Dynamics’ robotic dog Spot in data collection, the advances brought by AI in data analysis, the promise of foundation models of AI, and achieving real-time loss intervention through AI.

Segment 1: (28:24)  In the next segment, Pete and Matthew reflect on how the topics covered in season 1 of the podcast align with real world challenges, a growing emphasis on risk prevention, technological advances that aid risk management, the advantages of real-time data, the evolving landscape of insurance, the need for community and industry collaboration, demonstrating the ROI of adopting prevention tech, and the future of the predict and prevent model. They also field questions from our audience on predict and prevent.

Nancy Greco of IBM
Nancy Greco
Distinguished Engineer
IBM Research
Show Transcript

Pete Miller [00:00:03] Hi, I'm Pete Miller, president and CEO of The Institutes. We're a global not for profit whose mission is to educate, elevate and connect the people focused on risk management and insurance. You're listening to Predict & Prevent a podcast that explores how technology and resiliency can prevent losses before they occur. In each episode, we learn about innovative solutions and hear from leading experts on how they are making these approaches a reality today. Welcome to the finale of Predict & Prevent season one. It's been my absolute pleasure to lead this journey of conversations of experts and leaders sharing their predict and prevent strategies. To kick off this episode, I sat down with one such expert, Nancy Greco, a distinguished engineer at IBM. Nancy has been with IBM for nearly 40 years after getting her Bachelor of Science from Cornell University. Nancy has accomplished a lot in her tenure at IBM. After joining as a chemist in 1983, she moved on to developing semiconductors and managing manufacturing. Next, she joined the research department, where she created AI and Edge Computing Solutions, and she recently embarked on a new chapter with IBM, helping clients to explore ways technology can solve their biggest problem. Nancy and I discuss the benefits of edge computing and the richest of data that can be obtained through digitizing factory processes. Nancy and I also take some time to discuss IBM's partnership with Boston Dynamics to develop an AI powered robotic dog named Spot. Nancy shares why Spot can be a more effective tool for gathering sensor data than IoT. It can prevent problems, monitor potential risks and operate in situations that can be hazardous to humans. Let's hear from Nancy. So Nancy just saw it as a level set for everybody who's listening. Can you just give us an update on your career background and how you got into your current role at IBM? 


Nancy Greco [00:02:25] Well, let me start by saying my career path is a bit of an anomaly. So I came into IBM 40 years ago. I was creating photolithography so that we could print images on a silicon chip. And I didn't plan to stay at IBM. Right? I was young. I was going to see the world. But what was crazy about it, I got so fascinated in how we created those images. And about six months into the job I was in R&D, I got moved into something called manufacturing, which I really had no clue what it was. And so that moved in there. And, you know, we're manufacturing semiconductor chips 24/7. People thought I would be terrified. I loved it because I loved the drama of what's going on in a manufacturing line. I liked to solve puzzles, which were problems. And so I went from photolithography to I had every job imaginable in semiconductor, bringing up tools, quality analytics and all of that. So I did that for 25 years, and what I had to do is create a lot of my own analytics. It just wasn't there. My job was to keep tools up and running, make sure the quality was met. And so after 25 years, I said, You know, I've done every job here. I got recruited to come into research. And they said, Nancy, you've been on the receiving end of research and you've complained heavily about how we throw it over the wall. All right. You're going to come into research to make sure when we do hand it over the wall, it works. And so I started working, of course, on analytics and how we can improve it. And of course, that introduced me to AI. And then with AI, I started looking at edge computing. Why? Clients were saying, "Yes, you told us to go to the cloud, have you seen my bill?" And so with edge computing, the natural fear came in through robotics. And so that's kind of the short entry into my career. And I'll just conclude with that, like, yeah, I'm sure I could retire. But it's like the perfect time to be in technology. You've got clouds, everybody can get access. You've got tools and AI so that everybody can really unleash their creativity. And that's why we'll talk later about foundation models, because now everybody can come to the table and help solve a problem. 


Pete Miller [00:04:40] Yeah, that's really cool. And you know what, as you said, what a great time to be in in the career. Right. Can you just just quickly, for our listeners, just define edge computing? 


Nancy Greco [00:04:51] So traditionally, everybody kept the data in their factory and had numerous data centers, numerous servers, and it was very cumbersome to keep them up to date between the software and the hardware. So then there was a big move to go to the cloud, put everything up in the cloud. Well, one of the problems you're run into, it's kind of expensive. You're storing all your data in the cloud, you're doing all your compute in the cloud and oh my God, if you're sending data to the cloud, someone can intercept it and come back to you. So a new concept had come out many years ago called edge computing. And what it allows you to do is a couple of things. You can store your data in your factory with whatever footprint you want, but more importantly for me, what I could do is I'll create an AI model in the cloud because I have to send tons of data up there. I take advantage of the cloud in terms of its elasticity. I might need a thousand GPUs and TPUs to create my model. I'm not going to have that on prem. So once I create the model though, I can containerize it and move. So I create a visual analytics model in the cloud looking at all sorts of issues, you know, looking at environmental cracks, spills, you know, assets. Now I carry a very small containerized model. I put it on an edge server, which can be the size of a Raspberry Pi. And now the data, the raw data goes to that edge server, does the analysis and just gives me the result. I don't have to send the raw data to the cloud, and that's really important for two factors. One, the cost. It's costly to move data and then two security. If data is not leaving, no one can trace it back. 


Pete Miller [00:06:33] So it's kind of the best of both worlds, right? You got all this capability in the cloud if you want, but you can localize it and kind of get your job done inexpensively. Well, relatively inexpensively, closer to the edge, closer to you. So sure, can you tell us your view of how like risk data has progressed, like moving from one dimensional data to proliferation of sensors and everything we're seeing to, you know, to things that you see? 


Nancy Greco [00:07:01] Oh, of course. So I'll give you a, you know, real life experience in running a fab. So back in the day and we're talking 40 years ago, you did a lot of spreadsheets, right? And you would look at it and then we thought it was really cool to start doing time series analytics. Now, in that regard, you're still looking at one parameter at a time. And what I got very excited about are two trends that have happened, you know, recently, the introduction of sensory data. So I worked in semiconductors. You will not find any more instrumented machines than a semiconductor tool, whether it be a lethal tool or a reactive-ion etch. And those tools still went down and unpredictably. And in semiconductor you go down, you could be losing 100,000 dollars an hour. So you can imagine phone calls you get when your tool goes down and you don't know why. So I would go out there and it's like, my goodness, we just came up from a preventative maintenance. How could the tool go down? And there would be an operator arms folded. I knew, Nancy, it was going down. Like, why?It didn't sound right. I smelled something. It felt hot. And that was when the light bulb went off a few years ago. We're missing data. So what we're starting to do is use sensors, thermal imaging, vibration, accelerometers, all a sort of visual analytics. Look for cracks, look for spills, data that is missing today. But sensors aren't picking it up until it's catastrophically too late. So why that's very interesting for insurance is because we're monitoring assets this way and we're starting to digitize that data as well. So we're bringing that data. And today, I'll be honest, it's still pretty much, you know, model. So I'm still looking at one parameter at a time. We want to start merging that. So we're going to take like scattered data, which is all, you know, what's the pressure flow, etc.. We're going to start narrowing it with that sensor data to really get to understand the health of that asset. Now, the other thing I discovered when we came out of there or when we went into the pandemic, I should say, manufacturing was shut down. I'm like, you guys can run lights out. I knew that the level of automation had gotten to that ability that you can run lights out. No, we can't, Nancy. It's required that someone do a walk around to make sure everything's okay and that walk around as a human. And that's when I started investigating a lot of that walk around data is a clipboard, and in some cases our clients were showing 21 pages of clipboard and it's a check mark. And I said, What are you looking for? Things, you know, crack spills, this and that, and I said, But it's a checkmark. Right. And how do you go back and remember how big that crack was a day ago, a week ago? And they're like, Well, no. So what's really interesting now with these sensors. I'm going to digitize that data. How big was the crack on Monday? Did it get bigger on Thursday? And actually, from that, I can now predict when will it, what's the propagation of that crack and when is it going to become catastrophic? You can do that with spills and rust and vibration. So that's what's really exciting. Data, even today, most of our clients aren't collecting it other than a clipboard. We want to digitize that. When you think about that, how will that transform not only the ability to improve the asset performance, but understanding the environment. And this is why I think insurance companies are very interested, because I've gone through audits, you know, and go prove you did the inspection. Well, here's the clipboard. Nancy said she did it. You have you, how do you prove you did due diligence? Now, let's face it, catastrophes can happen. But at least if we showed we were checking out, we were collecting that data. And that data is valuable not only for compliance, but it's going to feed into A.I. Could I have predicted that the foundation was going to collapse. Right. And so that's where it's really exciting. We use the data from the industry and we start building models, of course, insurance can use to help prevent the problems from occurring. 


Pete Miller [00:11:14] So, Nancy, that's amazing, right? That's really cool, because you could imagine also, right, you go, oh, there's a crack. But I know from prior analytics that that crack, that's okay. I don't need to fix that. Right. And so I don't need to take the expense to fix that crack because I got enough analytics on this machine or machines like it to know that that's not a problem. I mean, that's kind of what I'm hearing you say as well, right? 


Nancy Greco [00:11:41] Exactly. So if you see a crack, a lot of our clients say, well, is it bad? I mean, we don't know. We're going to monitor over time. If the cracks been there for a couple of months and has it moved, you're going to be okay because you're exactly right there. It's going to come down to do you fix it now or how much time do you have. And you know what's interesting from IBM, we've done preventative maintenance for so long, it's the same decision, right? I've been in semiconductor. I had a big client. I had to get these troops out. Do I shut the tool down now because the PM schedule says I do, or can I get two more hours out of it so I can meet my production flow? It's the same thing. It's all about risk. And when I have data saying, Yeah, Nancy, you've got some room, you can run another two hours, you got a month and a half before that crack becomes catastrophic, and that's how you manage risk. 


Pete Miller [00:12:30] So I read a lot and I mean, we've talked a little before about IBM's partnership with Boston Dynamics. Can you talk to me about that and what advanced robotics play in this whole predict, prevent thing? 


Nancy Greco [00:12:45] Sure. And it's interesting when you talk about a career, how a phone call can change your career. So I was always a big fan of Boston Dynamics. The problem I had, I couldn't close the business plan for IBM. It's like, what are you gonna do with a robot, Nancy? And so two years ago, vice president of consulting, Skip Snyder called me and said, hey, you know, Nancy, you're working on AI and edge, is there something we can do with Boston Dynamics? In a flash, I saw that robot now can be my roaming edge device. He's like, what are you talking about? I said, we can load spot up with a lot of sensors and do the analytics literally on its back. And with that, he said, okay, what do you do tomorrow? You need to be in Boston where I have a partnership discussion. And that's how it all evolved. Now, the reason I was able to get there, I've been working in industry 4.0 and I've been doing, you know, POCs where I go in and I put a gazillion IoT devices in and some of it's funny stories. So I went out, fitted this one client with a bunch of acoustic sensors, you know, ten pumps. And for weeks I'm listening and listening and analyzing the data and acoustics generates a lot of data go into the cloud teaching, teaching. So I finally called the clients of, Man, I don't see anything. He goes, I mean, these things don't fail for what, maybe every five years. Okay. I'm going to be listening to this data 24/7 for five years before it detect a fail. That's not efficient. So what I can do is spot now to reduce data movement. So remember, data movement equals dollars. Right. So now instead of having 24/7 on all these pumps, the new pump I checked every two months. The very old pump based on the preventative maintenance, I might check every week. If I started to see a drift, I checked it every day. So now think of two things. One, I don't need all those IoT devices because a very common question I get from a client, which IoT device? Well, depends on what the failure is. So if you want to make sure it doesn't fail, you're going to need a thermal imaging device, an acoustic device, a camera, an accelerometer, all on one machine. Why have a thousand machines? Chaching, chaching, chaching. It's not only the cost of the IoT device. You've got to drop power to it. Right? That's not easy in manufacturing and also to all that data, we were jamming the network and so we just had too much data. So now on Spot, I can decide when to go to the machine. I can change it on the fly based on data. Even this new tool starting to drive. Let's check it every day. So I reduce the amount of data and now the sensors can be plug and play. Maybe thermal imaging wasn't the best way to detect it. So that's what's really exciting. Again, giving clients choices. You know what sensors will detect the problem? I can load Spot up and say, Hey, the best way to detect this is a vibration monitor. 


Pete Miller [00:15:43] So, Nancy, just us just to spell that out a little. So then theoretically, as I understand what you're saying, you could say, well, pumps, the best detection device is a camera. And so on Tuesdays, I'm going to check all the pumps or the pumps that because of their age or whatever I need to check and I'm going to load Spot up with that and go do that on Tuesdays and Wednesdays. You know, for fans, the best is a vibration sensor. And I'm going to, is that kind of what you're saying? 


Nancy Greco [00:16:10] Well, actually, we're doing that right now, so we have to each our own food or put it that way and I should think dogfood. This Spot looks like a dog. So what we're doing in IBM's industrial waste center, part of our manufacturing site in Poughkeepsie, I have Spot doing that. So what we're doing right now, we have pumps and there is multiple components on a pump, right? I have the fan, the casing, the motor, the inlet valves and all that. So what we decided is to monitor that through thermal imaging first. So Spot has a thermal imaging on it. And what we do, we don't say, Hey, the pumps hot. We actually segment is that the pump, the inlet valve, the casing or the fan and we're creating that behavior map of those pumps over time. This is data they didn't have. They literally used to go put their hand on. It feels kind of hot. What if your hands are callus? Well, it doesn't feel as hot. Does it not feel as hot because your hands calloused or because there's a problem? Ambient temperature really makes a difference. We're reading analog gauges next to the pumps. So we're taking the temperature and we're reading the analog gauge. What's the chemical coming in? What's the volume? What's the pressure? And we're watching the heat signature. So you can imagine what you can start telling on that. 


Pete Miller [00:17:27] That's amazing. That's pretty cool. I didn't realize you could read analog gauges. 


Nancy Greco [00:17:33] We did teach Spot through an AI pipeline we created to do that. And a lot of people say, oh, Nancy, analog gauges are out. No, they're actually growing five to ten percent. Why? There's a lot of explosive material out there. You're not going to be allowed to put a digital gauge on a hydrogen clank unless you want to be in the next neighborhood. 


Pete Miller [00:17:55] You talk about AI and interplay with robotics and things like that. So, you know, we know I think that AI is pretty essential for making sense of risk data, right? So how do you see AI's analytical capabilities improving from where they are today and in fact evolving? 


Nancy Greco [00:18:16] So I want to look at it in two different ways, because AI is approaching parametric data, right? Pressure and sensory data and starting to aggregate that start moving that to a multimodal model. So what I mean by that, so what does a human do today when you hear something? You hear it, you go look at it, you touch it, you smell, maybe taste it. I mean, who knows? And we do that all instantly. So we do that as what we call a multimodal. You hear a smell, look, taste. And you're going to start seeing the sensors come in to that as well, because it helps to validate it. Right. If we're seeing a signature, you know, you want to be able to validate it. Is the leak due to a vibration? Is there a smell? Is this thing overheating? So that's going to start giving holistic approaches to how to look at the data, so this parametric data and sensory data. The other excitement which everybody is seeing with OpenAI. So ChatGPT is how we're dealing with language. So what is kind of interesting, a large part of the data out there is all human text. People write reports, you know, work orders, what did you do, how did they do it, what is your observation, think of a clipboard. So what's really fascinating about foundation models and that's what Chat GPT is, and IBM just announced our version of what's next. The big difference is we're going to be industry focused. You know, ChatGPT is wonderful for consumers right now, but we're going to really focus on our clients and what do they need to solve. So what the language model, they call them, large language models, are going to start to look at all this data. It'll look at your warranty and we'll look at your manual. It's going to look at the history of the tool, you know, any work order your PMS, again, all in a language right now and start to look at nuances so that you can understand why did it fail, right. What we were missing. And so a lot of people get confused. They say, Can't you do that today? So let me give you kind of an example of what's changing with foundation models. Today, you know, even using an NLP interface, you can say, what are my five worst performing tools? And it's going to go out, grab data out of a table, and it's going to list them. And it might say what type of tools. It's what a foundation model is, what targeting what's next to do is you can say, why? What are the attributes of my five worst performing tools? So it's going to look at a collection of data. It might be at a certain age with this warranty, with this type of workload and maintenance schedule, and now you're going to get a richer sense of how do I manage these assets? So I'm really excited about the potential of what foundation models can do because it's going to take AI and put it on steroids. And that to me, that's going to be really, really fascinating. 


Pete Miller [00:21:12] It's a fascinating time. I built my first AI model in LISP in like 1989, Nancy. And so to see the evolution and in particular, how they've taken off, particularly the generative AI models, is fascinating to me. Are there any other new innovations or developments that you think will be game changers for insurance and risk management, specifically as we try to implement more predict and prevent capabilities? 


Nancy Greco [00:21:47] Yes, I do. And it kind of touches on themes we kind of talked about. We're missing a lot of data. Buildings are collapsing. We did know why. Actually, just recently with the storm that hit us, my area flooded. And one of the reasons is the easement that's coming up to major highways did not do the assessment correctly. And so there's flooding our plains because all this water is coming in. So the first thing is we're not gathering enough data. And so we're going to be pushing even more IoT devices. And when you have a very macro area like I got to monitor a highway, how much water is building up. So they're going to be an evolution of more IoT devices and then taking that data to try to understand. And now we've got to start bringing in weather, you know, weather, weather predictions and patterns. Those are dramatically changing. And then also, you're going to have to start to pull in this civil infrastructure. What is there? Why is it flooding? And it's going to be this aggregation of just massive amount of data. So again, using foundation models or a representation called knowledge graphs, how do we help the humans see what's going on?  Because with the flooding that we've been having, they keep saying, well, it shouldn't do that. Well I have two feet of water in my driveway. Don't tell me it shouldn't have done that. Right. And you can see the easement pipe. So in terms of preventable losses, there's a lot of data. I mean, how we've got to understand wildfires, we've got to understand local flooding. We've got to understand buildings and construction integrity. And that's what we really got to start pushing because we've got to make sure we understand why these things are happening. And then we're going to back up and build the preventative. And I do think things like foundation models are going to show us that nuance. So what we're going to do today, and it's starting already. It's on the road map. Foundation models are primarily language, but now we're going to be doing visuals, right? And then we'll bring in the rest of the sensor data and then all the parametric data. Now you can really start asking questions. I think, in a small amount of time, in a few years, you can say why did that will occur? And it's going to give me all the factors with this amount of rain with that is, you know, and here's how you would need to fix it. And I think that's what's going to be fascinating. The same thing with wildfires. You know, what do we have to do differently? We can't just keep watching it occur. 


Pete Miller [00:24:17] I think increasingly, you know, where the industry is seeing you know, there's a real value add to this predict and prevent. Right. Because the best loss is the one that doesn't happen, as you say, for everybody. Right? That's a win win for everybody. And if you can bring both up as an insurance organization, if you can bring pre loss mitigation in post loss recovery, that's the ticket. Right. And that is I think more and more organizations are seeing that. And that's what we're trying to kind of spread the word on. 


Nancy Greco [00:24:49] I mean, I don't know if you saw West Point, three inches of rain an hour. We have to start to anticipate that. It's going to get worse. We've got to start doing assessments. How do we handle the climate change? And I think that's when insurance is biggest concern. We're flooding where we never flooded before. West Point, they're calling it the thousand year storm. Well, that wasn’t really predictable. You know, the weather patterns are getting harder and harder to predict. But I think also with the insurance companies, we can start saying how do we build infrastructure better, cannot withstand flooding or extreme temperatures or winds. And that's where I think it would be nice to have a coalition come together. We're going to have to build smarter and I don't think we're focusing on that right now. The analytics are there. How do we build smarter? 


Pete Miller [00:25:38] The analytics, the capabilities, and as you said, the real time intervention capabilities. Right? That's what excites me, Nancy, because I'm like, well, wait a sec, there's enough data and it can be timely enough and the decisions can be made quick enough that you can avoid the problem in real time. Then that I think is new, right? As opposed to building codes and all that great stuff. That's awesome. And we need to bump those up. But the real time intervention based on data input and, and, and analytics leading to an AI system that can make a decision like that and boom, not a bad thing just was avoided. 


Nancy Greco [00:26:20] And what's really critical about doing it now, like we should have done this yesterday. If we do need new materials to withstand what the climate change is going to introduce, we need to know that right now. And as you're seeing with quantum. Quantum's getting to the point it can understand molecular behavior. So now we're going to have to start saying what is the right metallurgy to use or what is the right word or to be honest, what trees should we have in these wars? They can't just keep burning. Yeah, it'd be nice to keep them, you know, indigenous, but it doesn't work anymore. We're going to have to do a blend. What can withstand the temperature? And so it's almost be a call for action. We have to know what's causing it, and then we have to kick off. We might need new materials, new ways of thinking, new ways to manage us because we can't keep just fixing it. We have to truly prevent it. And that's going to require a different thinking. 


Pete Miller [00:27:17] Well, Nancy, thank you so much. Thank you for your work. Thank you for your insights. And very, very appreciative of your time today. Thanks again. 


Nancy Greco [00:27:25] I thank you for letting me talk to your community. Because it's just I can't tell you how fascinating it is. And everybody, to whatever degree get involved because it's a whole lot of fun. It's just absolutely fascinating. 


Pete Miller [00:27:42] I really enjoyed my conversation with Nancy. It's truly inspiring to hear about the incredible work she's doing at IBM. And I especially loved hearing about Spot. It sounds like this technology is not only leveled up data collection, but cost savings as well. Next up is a truly exciting segment with my colleague Matthew Kahn, president of Risk & Insurance, and one of the producers of this podcast. Matt and I discussed the highlights from this past season of Predict & Prevent, our takeaways and predictions for what's next in risk management and insurance. We also answered questions from some of our listeners on how to apply predict and prevent. 


Matt Kahn [00:28:24] Well, hello, Pete. It's exciting to get a chance to step behind the curtain and into the driver's seat for the final segment of the final episode of our first season of Predict & Prevent. 


Pete Miller [00:28:37] Yeah Matt. And I'm very excited. It's been a great season. I mean, the work that you've done and the folks have done to make this happen has been outstanding in my view. So it's, you know, I sit in front of the mic, but it's the work that everybody else has done. And you know, that's kind of I think, led to what's pretty interesting podcasts, pretty interesting topics. 


Matt Kahn [00:28:59] Absolutely. It's been really rewarding, particularly as we're going to do today in a little bit talk about some questions that we got from our listenership, which is pretty exciting. So we've gotten a lot of feedback and that's really been something which is rewarding to know that, you know, we're aligned with our audience, but also that this clearly is a topic which is very important and in fact is growing in importance. Looking back, it's really amazing. We started this podcast in November of 22 and since then, you know our premise that the risk side is getting so much more severe that we really need to focus on preventing losses in the first place for the benefit of society as well as risk management and insurance, and at the same time utilizing technology, which is also growing to make that happen. And clearly over the last six months, just all the events on the risk side, climate driven and otherwise, and with what's happening in several states, insurance markets. And then when we started this the words ChatGPT weren't even on anyone's tongue. And now look where we've come in six months. So it really is amazing just to see not only the reaction to the podcast so far, but just how much more relevant it's all been. 


Pete Miller [00:30:21] Yeah, really. I mean, as you say, on both sides, right? Like I was, you know, just earlier today looking at these pictures of Maui now, like, unbelievable. Well, Lahaina is a nice little town or it used to be. Now, it's 85 percent burned down. Right. So, you know, think about that. And as you say, California wildfires, what's happening in that market. Florida, you know, what's going on with the industry down there. And, you know, and you know, we've said a lot in this podcast that now is a real inflection point because, yeah, risks are getting greater for sure for a variety of reasons, but the capabilities and technologies are also getting better. You know, Matt, I know you're very interested in AI and I am too. And you know, I always say that. I built my first AI system in 1989, and that was the dinosaur era. And it was extremely frustrating what it can do now. And combined with the amount of data that we have available, this just a whole new world. And it's very exciting, as you say, even in the last six months, all with the aim of making people's lives better and safer. And that keeps me very excited because that's the promise I think of sort of the the evolution of this industry. 


Matt Kahn [00:31:50] And while AI has some risks, I think it really is can be hopeful in the context of predict and prevent about how it can really help us handle some of the very present risks that we've talked about all season and you just mentioned with Maui. So actually this is a great segue into our first, Ask me Anything portion because we've got a great question, Pete from Emily in Chicago. And she actually asked how you see generative AI being used in the insurance or risk management communities. 


Pete Miller [00:32:25] Well, Emily, I think generative AI obviously is a really a revolutionary technology. It can be used in a number of different ways. So if you think about what it does, right, is it takes data and it synthesizes it and it presents it back to you in a digestible way. Now, Matt correctly says that there, you know, we have to be careful, but if you think about fraud detection, we can, you know, use that to spot patterns. If you think about policy comparisons or policy analysis or if you think that you're in claims and you have a claims file and you want to take all these disparate pieces of information of different types and distill them down to a summary, that's what that's what a large language ChatGTP like model can do. Think about customer service. If I'm an insurer, an insurance company, through a series of prompts, I can make sure that my policyholders, when they call my organization, get a much more complete and standard, if you will, predictable form and useful form of customer service. 


Matt Kahn [00:33:45] It could be like even like a chat bot that homeowners could use to like if a disaster is coming and ask it what, you know, what to do or things specific to their policy. 


Pete Miller [00:33:56] If you're a small business owner in the Midwest and you know you're you have a farm implement shop, well, what risk do you face? Well, you can figure that out, right? You could put in a profile of who you are. It could go out and assess your risk and then kind of build your risk profile in a generative AI way. So that's just one small example. Certainly if you're a company and you're looking for fraud detection, it could, it could go out and it could read patterns and not only tell you existing patterns of fraud, but generate, you know, things that it's seeing that are potential patterns of fraud. Right. So it can anticipate future problems and then create strategies to avoid those. So there's a lot of things that that can be done on, certainly on predict and prevent as it relates to AI in general and more specifically, generative AI. 


Matt Kahn [00:35:05] Our next question is from David in Houston. And David's interested in sectors that you think might be particularly well-positioned to benefit from using real time data to prevent losses. So are there any particular sectors you think that some of the things you've talked about can apply to? 


Pete Miller [00:35:28] Sure. Some of the sectors are kind of doing that already, right? Like personal lines auto crash avoidance systems. Those are real time systems. Read your driving habits and what you're doing. And if you're going to cross a lane, it's going to stop you. Some other personal lines, I think, are a little more difficult, only because there's a lot of retrofit involved. You know IBHS talks a lot about, you know, how old the housing stock is in the United States and you'd have to do a lot of retrofit. But on ones, you know, commercial lines, I think there's some really good opportunities. We interviewed we interviewed IBM and Spot the robotic dog. And I thought that was fascinating because rather than installing sensors all over the place, Spot goes around with sensors. So it's easy to install. In fact, you just kind of turn Spot loose and goes around. I think, you know, worker's comp, we interviewed, you know, StrongArm and we saw that product. I think on the commercial line side as well. It's easier to calculate ROI, the business rationale. Right. A little more difficult perhaps with some homeowners. But on commercial lines, you know, you can really say, look, you know, as we did with the water sensors, you know, the property manager that we interviewed said, Look, you know, you know, I know what a water loss costs me and I know what these sensors are. And, you know, I can do that math. And it makes a good business decision to do that. So commercial lines, commercial property, workers comp, I think those are areas where the you know, it could be adopted pretty quickly. I certainly think there's many on personal lines, including, you know, we interviewed the CEO of Ting and things like that. You know, so I think commercial lines is is probably a little more ripe for some innovation just because of the ease of dealing with the business models. And but I do think there's opportunities across the whole spectrum. 


Matt Kahn [00:37:47] So that's a great segue into another question we got from Sarah, who's from Miami, and she was wanting to know about how predict and prevent can apply to smaller businesses. Is there any particular things that a smaller company should be thinking about? 


Pete Miller [00:38:04] Yeah, So starting sort of at the general high level, I think every business is really an exercise in managing risk. When we talk about any small business, there's a lot of different kinds of businesses, but most small businesses are entrepreneurs. And you know what, what risk would an entrepreneur face? So you don't want to do anything, obviously, that risk your own assets or any more than you have to being an entrepreneur. So I think, you know, I would look at what risk do I face in my business? How do I manage those risks? That's to me the very first thing. And then say, well, let me try to quantify the impact of those risks on my business and then develop an understanding which ones can I prevent and how can I prevent them. You could get there's a lot of resources to help you do that, but it's kind of a stepwise process. And then those that I can prevent and let me adopt preventative measures in order to do that and then those that I can't, then let me try to find some kind of an insurance tool. So I kind of, you know, kind of go through that stepwise process. You know, if you if you employ that, then I think you can get a good understanding of what you're facing and a good understanding of how to manage them. 


Matt Kahn [00:39:27] So switching gears a little bit to a topic which is very close to our heart, a mission. Kevin from Seattle is asking how you see the role of the risk management profession evolving if predict and prevent becomes something more, more of a priority to more businesses. 


Pete Miller [00:39:49] Yeah. So this is very near and dear to my heart and to our heart. So, you know, if you look at the history of risk management, it's generally been thought of as, you know, detect and repair. So we've talked about that in the past. But the industry really started out as something bad happened. How do we make people whole? And that's a good model is then the industry has done fantastic work in helping people get whole. But it's premised on, you know, we're going to try to price. Think bad things that have happened before and charge a premium so that we can, you know, repay people when something bad happens. Where I get excited about this and where I think it's going to change the risk management industry is we put a front end to that equation and what it says is, well, Hey, the technologies and the data and the tools are available now where we can go to our clients as risk management professionals and say, Guess what? Now we have tools to avoid the loss in the first place. So whether it be displacement from your home or displacement from your business or lost business, we can start to help as a risk management industry. We can start to help people avoid those in the first place. And that to me is extremely valuable. So yes, we have a backstop. Yes, we will provide tools and products and services if a loss happens because they're always going to happen. But I think the mindset of the industry will now be, well, you know what? Let's look at this holistically and let's talk about how do we avoid accidents and losses in the first place, as well as how do we help you recover from a loss? And that is a fundamental change. But I believe it's a change that this industry can really undertake and add value at the end of the day, you know, if we do that. 


Matt Kahn [00:42:06] We also got several questions in response to episode four, which was the one that Frank hosted and featured to regulators and talking about the regulatory impact and what it can do to help this transition for risk management, the insurance industry. I want to be sure I'm giving our listeners credit. So a couple of these came from Jessica, who's from Boston and Chris from San Francisco. But in general, the theme to these questions really focused on how do you see or how would you describe the interaction and relationships of how communities and insurers and the technology and regulators, you know, how do all these groups work together to try and make this more of a reality? 


Pete Miller [00:42:57] Well, I think this is a very key question in order to get this implemented right. So it's going to take all of us and it's a community. And I think, you know, as you said, there's various members, members in that community, you know, and I think, for example, you know, there's have to be community groups or local civics that are going to have to say, well, you know what? There's a lot of wild grass growing right next to the buildings and homes, and it's in a common area. And we're going to have to make sure that's mowed. Right. That's one small example. I use that example, Matt, because my mother actually owned a home near the Marshall Fire in Boulder, Colorado, if you recall, that fire which spread very quickly similar to the campfire, although not as destructive and similar to these fires in Maui. And part of the problem was the city would not cut grass and it spread the wildfire. So that's one thing. I think regulators, it's a heavily regulated industry, right? So there are some things around discounts and rebates and things like that that regulators are going to have to go, yeah, you know what, this is, you know, we're going to allow some of these things because they make sense in the context of helping make people's lives better and safer. That'll be part of it. I think businesses and policyholders, you know, they're going to have to say, you know what? This is how I want my insurer to approach me and say, you know what, I really don't want to have a loss and how can you help me not have a loss? And I think it's it's part and parcel to an insurance organization saying here's products and services that I can offer. So it's a change. It's you know, this industry is 350 years old but I think it is going to take all of us. And regulators are a key part, right? Properly so, you know, regulators have a hugely important part to play in our industry because, you know, we don't want insurance companies going bankrupt and people's claims not being paid. It's also true that regulators are going to have to embrace innovation. If you look at it, they're generally receptive to it because they use a principles based approach. So I think that it is going to take all of us, each of us has our part to play regulators being a very important part. So I think that's crucial in order to make this kind of mind shift. 


Matt Kahn [00:45:30] Our next question comes from Jennifer in New York City, and she's interested in hearing some details about how to make the business case to her employer to justify the investment and expense for some of these technologies. 


Pete Miller [00:45:49] Well, thanks, Jennifer, for that question, because that's a really great question and obviously very important to the things that we try to do. So I would point, you know, in one of our episodes, we talked with George Chedraoui. And what George does is he manages commercial office space. And this is a prime example to me, because George said his solution, working with his insurer, revolved around how to mitigate water damage in the case of a flood in one of his buildings. He estimated that the loss from a flood was around 100,000 dollars. To install the sensors was less than 5,000 dollars. So that's a pretty good ROI, right? You can think of others just in terms of lost days StrongArm that we talked about, that the device that warehouse workers can put on saves hundreds of injuries each year because it warns people in real time if they're doing improper lifts either from over their head or from bending over. And so that saves a lot of folks from getting sick and from, you know, from the businesses point of view, there's, you know, many fewer lost days and they don't have a long tail medical payout on their policy. So, you know, I think this is an area that needs to continue to be developed. But I certainly think from the policyholders perspective, there are good exemplars of a business case that can be developed and explained to their boss. 


Matt Kahn [00:47:38] Our next question comes from Charlotte in Atlanta. She is interested in hearing your thoughts around prevention and what were some of the and where do you see as the most impactful approaches to prevention with predict and prevent? 


Pete Miller [00:47:56] Yeah, Thanks, Charlotte. So I think there were several outstanding examples throughout the podcast. One in particular addressed a problem that we see in the press today California Wildfires and Anukool Lakhina from BurnBot had developed a device and what does it do? It recognizes that, for example, in the campfire that went through Paradise, California, a few years ago, it is, I think it killed 85 people and destroyed 90 percent of the buildings. You know, an analysis after the fact said that there were specific spots where if they remove some of the fuel, dried grasses, dried, you know, dead trees, that there was a very specific pattern that if they had a remove that fuels that fire would have stopped and not spread and then using his device Burn Bot. And what BurnBot does is in a clean, environmentally safe way, could go to those places and burn off that fuel without starting a wildfire and then remove that risk so that perfect predict and prevent. Right. It said, what's the risk we're facing? It's fuel, unburned fuel in a particular area that if it ignites, would cause a severe, severe loss. Predict and prevent would say, well, why don't we analyze that and then take a predictive measure in order to remove, you know, a bad accelerant in this case. So that to me is something where, you know, a predict and prevent mentality said, Well, let's do the analysis. Let's understand how this spread and let's take and remove an accelerant to these fires. Well, you could easily apply that elsewhere. Right. So that's kind of the mindset, the risk mitigation mindset. I think that predict and prevent is trying to get folks to adopt. 


Matt Kahn [00:50:07] Pete, I think we have more questions, but that's about all the time we're going to have. I just want to say it's been a real pleasure working together. I think this is a really important topic and look forward to helping to bring the discussion further over the next several months. 


Pete Miller [00:50:24] Well, Matt, I hope you know how much I've enjoyed working with you and with the other folks that have put this together. This is a fascinating thing for me. And I feel really grateful and really looking forward to the next episode of Predict & Prevent. And that brings us to a close on season one of Predict & Prevent. I'd like to thank all our guests for coming on the show and everyone who tuned in to the podcast. I know I learned a lot from the expert we spoke to, and I hope you did too. Stay tuned for new episodes and season two of predicting Prevent where we will continue to highlight exciting innovations that make our world safer. Predict & Prevent is a podcast brought to you by The Institutes. Subscribe on your preferred listening platform and join us for future episodes where we continue to dig into this approach and the opportunities that come with it.