Go back

Mark Miller, Founder at Insurevision.Ai - PIR Ep. 876

from Profiles in Risk

38m 42s

Mark Miller, Founder at Insurevision.Ai - PIR Ep. 876

Mark Miller of InsurVision.ai shares an update on his company’s groundbreaking Fleet Vision product, launched at CES 2025. The system leverages AI and machine learning to analyze dash cam video data from fleets in real time, providing insurers with precise, up-to-date risk insights. Unlike traditional dash cams that miss up to 60% of contact events, Fleet Vision achieves near 100% accuracy in detecting crashes and estimating damage—enabling instant, reliable FNOL (First Notice of Loss) processing. The solution doesn’t require deep integration with underwriting systems, instead offering a user-friendly portal where insurers can review risk by driver, fleet, or quarter. It runs efficiently on modern edge chips like NVIDIA Jetson and Rockchip, allowing real-time risk prediction and crash prevention before incidents occur. This shift from reactive to proactive safety reduces false positives, improves driver accountability, and helps fleets reduce risk with minimal staffing. Priced on a monthly subscription model, it aligns with existing telematics pricing norms and is designed for scalability across large commercial fleets. With over 3,500 experiments behind it, the technology offers a deeper, more accurate understanding of driver behavior and fleet performance—transforming how insurers assess and manage risk in the commercial auto space. The future focus is on expanding edge-based, real-time risk prevention, marking a significant leap in automotive safety innovation.

Transcription

5973 Words, 31289 Characters

English
Hello and welcome to Profus in Risk. This is your sleep deprived and over the caffeinated host Tony Kennis back from two days of trade show magic in a non-intrus conference. First one in my history. This was amazing Today I have with me Mark Miller founder and CEO at insurvision.ai returning guest So first of all Mark, thank you for joining me today. How's he going? Agreed, very good. Congratulations on the magic show. I guess in America you never got Paul Daniels I'm not familiar who is Paul Daniels Well, I think I have to tell you Paul Daniels certainly for Anybody of a certain age was the TV magician that we grew up with in England in the 1980s. I sadly no longer with us I think I've read about him and you're correct Number one, I grew up in Costa Rica and yeah, most of the TV we got we had BBC on cable, but most of the TV we got was was American And so I grew up in the 80s. I was born in the two. I grew up watching copperfield and yeah, and a lot of people a lot of of magicians a little bit older than me grew up watching Mark Wilson and Dog Henning But yes, you are correct. I wonder you mentioned we we didn't so now I know we're near being a TV magician But this was really cool because I've done a lot of magic at insurance conferences I've done a lot of parties and I do magic on the street here in Atlanta, but For the first time this was I was hired as a some magician by a talent agency for two days To represent a really cool company and and I show up and and they told me it was it was a it was a private jets conference And I show up how it was so they told me it was it was the national the the agent told me was it was the National Business Aviation Association I'm like cool private jets cool. I can do this right. I show up. It wasn't it was very calm I had no idea what vertical was vertical is the largest helicopter conference in the world Okay, I cannot even describe and I'll post pictures on LinkedIn in the next couple days How cool this was because I I I did my magic and then I got to walk around and get pictures with like 50 helicopters of old types I know Awesome police. It was awesome. So so so Mark was on the podcast on episode 668 about a little bit less than a year ago I just give you an introduction to Inturbation and I for those who haven't watched it I will include that in the show notes and he's back today to give an update because things have evolved advanced To mark normally I say the elevator pitch, but in this case, so yeah, give tell us a little bit about Inturbation and while you're back So I think it was about a year ago um, we spoke um And I guess we probably talked about turning video making video useful to send a risk Um, if that needed to talk about that and we done since then About three and a half thousand experiments and we built a perfected video risk understanding system Which I want to tell you about maybe show you some of it Some uh show you a demo of what that looks like. So as a company We were a busy on that we launched the the product the solution the first you know major solution components is called uh fleet vision um Let's see. Yes, and um What all those figures at so not I'd like an insurance or a risk conference at CES at the largest consumable electronic show Yeah, well look at that if we have the system ready um very very very perfected late late late November December if you look to a um Fried show handbook You see that CES is on the 6th of January So no time like the present um that is spent two spent two years making everyone to get it out there So that's fantastic But at CES at CES at the 6th of January the belagio ballroom I forget on the ground floor um With um some of the guys that uh we we we launched it uh something which is called Kavisa which is a connected car um group and um Yeah, so it's been life uh I know a few weeks a couple of months. So I thought it was a good time to cash out that is fantastic So congratulations on on the product actually being live Congratulations on hitting cbs. How was the reception at cbs? At CES. Yeah, it was very positive. I mean We we focus on sort of a lot of verticals and one of them is The ability to run this at the edge in vehicles and so connected cars are kind of logical or logical place to actually to to to talk about a service like this We have a bunch of people um insurance executives, automotive executives um this is very good discussions As cs anyways, it's really interesting if if anybody that's never been I've never it's just such a Well, it's just worth going. I know that you you go annually to the ITC one and um and I got a bigger it's like seven times a year I'd love to see ones It's it's it's a big you know, it's a big trade show for sure. It feels Feels huge compared to something say that we might have in in England um, but cs is is is you know 20 times bigger Yeah, so it's just really it's just for anybody that's interested in technology I think it's you should go at least once um It's just interesting a quick parenthesis for for any listeners Any insure techs that are planning to go to cbs because you do more multiple verticals You will get a sweetheart deal with this magician if you bring me to see to cs as as you're uh as you're with magician Because I would love to do cs anyway continue Note it cs 2027 look out I'll bring the rabbits That's what Paul Daniel's used to be He used to do rabbits out of top hats Top hats yes Fire a little bit Carred to ropes. Yes. Uh, no live animals my insurance doesn't cover it. Oh, I see It's a mile since I watch maybe live animals a lesson. Yeah Yeah, they they they really uh, no, my insurance does cover. I joke about that. No, the problem with live animals at two two problems Uh, no number one uh, like the rabbit out of the hat is it's an easy trick. The doves are harder Uh, but the rabbit out of the hat is it's an easy trick The the problem or the out of the hat is that you get 30 seconds of entertaining and then you need to take care of our rabbit for 24 hours Uh, we don't have pets my girlfriend would kill me uh and and uh number two problem is uh, audiences have gotten really sensitive to The welfare of the animals which I agree with In like the golden age of magic They were not particularly nice to the animals like lots of lots of doves and rabbits died and sometimes in front of the odds Right, sometimes you pull it in any and it was dead, right and that today would get you canceled for life kind of thing So the poor those two reasons I I wouldn't do animals That makes sense, right? I'm re I'm re I'm re calibrating Paul Daniels as we speak Like like back in back in there and in his day Uh, and there are people they're they're like 75 year olds who still do do do shows and they do them really well And they've done them for forever and they're very good at thinking care of the birds uh, yeah, it's not it's not something I'm going to pursue Anyway for something for 27 CS for seven january straight off the new year perfect so so mark now that it's life tell us tell us tell us about it and and yet show it show us Okay, well, I got a sort of demo which I'd like to show you about The sort of thing an insurance company can do with this technology um, so I want you to um, um, I guess put yourself in the mindset and you've got some insurance experience Let's suppose you're an MGA or a commercial insurer You ensure fleets um You're gonna you're gonna request Let's say you're interested in and expanding your book of business or fleet purchase you You're gonna basically ask them for you know, they're lost lost runs for the last five years you're gonna you're gonna want some of the vehicle types Um some registration details stuff like that um And and and pretty much That a few other things it's gonna basically how you're gonna price price that risk okay, so No, obviously there's something in that because let's say the most predictive element of that is past performance like lost runs Now You know past past performance is an indicator of future performance generally speaking, but let's let's take the example where Um, I don't know something something significant happened to the fleet or it changed ownership or actually You know, driver shortages is a big problem. They have a lot of good drivers and they left and the new drivers that came in are not so good and so you could imagine over over a 12 month period. You know that maybe come some quite significant changes that might not be entirely obvious given the. Infrequency front clear flames so the fleet has changed. In a positive or negative way, but you're going to really understand it based on its last last few years of loss performance and so. What could you do about that? I mean, so what we have built as a system, which I'll pop up on the screen a second. Let's say you've got a most most most fleets in North America how the dash cam system of some description and that dash cam system will be logging. What are called dash cam event clips to the cloud. You can plug in our system fleet vision to the cloud for that in short set for that that fleet and extract all of the four with the road facing road facing event clips for the last 12 months and that will tell you two things so we'll tell you the total magnitude of danger that the fleet was exposed to in the prior 12 months. And we'll also tell you it's mostly my driver. So I'm going to pull this off on screen to show you what the place and files. So just if you just give me a second. And while you prepare that just while you're explaining it so I am a former underwriter. I'm a former middle marketer and writer. I didn't do commercial auto but but I mean, I didn't expect anything commercial auto. We definitely had commercial auto, but we weren't doing like truckers. But and I used to this I used to dispatch truckers before my insurance life. I don't know if you knew that I was I was a fleet manager. I was a fleet manager. My first job out of college. I was a fleet manager for three years flatbed trucks basically when I went on my drivers. I had 40 drivers company of 3000 drivers. I managed 40 and they were all over the lower 48. At that time, no camera. This was 2006 through 2009. No camera. We did have a little poll community that I could message them on and other phones. And we had like even the tracking at that time. I only I only got a ping from where they were once an hour, which was not great. So. Yeah, we had simply no way like I wasn't in insurance yet, but knowing what I know about insurance. How on earth do you underwrite that the right. Historically, the only way to do it is by who like your August three who wore your drivers and what what is their history. But it's yeah, it's just not it's it's it's very proxy now. Something like what you described to show us again, I'm not a tricky underwriter, but but holy crap like if we're considering writing, right, if I'm an underwriter and I get a submission and they have a bunch of power units and they have good loss history. Or and they can show me right, they can let me dig into their that's come data with something like this. Yeah, I'm going to be more comfortable writing the risk or on the other hand, maybe maybe yeah, yeah, I can totally see how this makes a big difference. Anyway, go ahead. So is the screen, the screen should be shared OK, you can see that this is, this is, I mean, this is a, this is a, sadly, this is a fleet that actually was put out of business by this by this crash, which is at the top here. And you can see that. Essentially, the fleet have three crashes in April, I think this is a European date format showing April, a March and December. That's the most serious crash. But the damage on this one is lighter and lighter. And you can also see all of the other events that the, um, that the fleet has been evolved with. So. I'm just going to play you one of those video clips, what it looks like. So first of all, I know, I mean, sure, the loss history was showing me that there were three crashes, but I can actually watch the three crashes. I don't need to watch any other videos. You know, there are hundreds in a lot of light thousands, even hundreds of thousands of video clips and this system will serve me up the three crashes that happened right at the top. So I can see them. So let's just play one of those. I'm just going to. Quick stuff. I'm going to put one of them. I'm going to play in the media plaque is a little easier to control on the web app. So got it. I'm talking about now. All right. This is the, uh, this is the worst rated incident, but took part to that fleet in the prior, yeah, but don't you get what? Do you see that OK? Yes, I can I can see your screen up. Yeah, because I'm just going to play the video. So sometimes, you know, media player and screen because it's a bit jerky, but, you know, it's smoother than my screen. So. Well, what we can see here is, um, we can see that there's this big spike of risk. You can see in clothes, it democated to do the top of the screen. So I'm just going to play this word for that. And it's, it's, um, I'm just going to go back here. He's about to overturn his name. Yeah, he's going to, he's going to, he's going to, he's going to roll it up. Um, you can see there's an imperfection on the road on the side of the road in the cup, which is a corporate wheel. And we're also going, we're going to too fast around the, around the time. OK. You can see the, um, then we hit the right and then we go. All right. And that's a role of a, um, that actually unfortunate put this way of business. Um, you can see it's a, it's a very serious crash, damage scores is a force in a hundred. It's, uh, it's just a terrible thing. And I, I, I, I, I am very curious, uh, was it a small fleet where, were they not properly insured like having a single roll over to try to business makes me wonder your insurance wasn't correctly set up. Yeah, the, it was a reasonable size fleet. It was about 25 cos eights. And, um, it wasn't the set up in, margins have been skinny. Um, difficult market. Um, and the impact of that on the insurance cost is going forward. Just push the business from like, you know, be a bit borderline to being over the edge. And it was no longer viable. So it's not that there wasn't insurance coverage or that it's, it's just that, you know, show this insurance coverage. But, you know, those costs they, if they're going to come through one more another to you. As I don't know, as a fleet operator. So I'm just showing the second left as well. I don't play with them, but, um, put this on the screen. Yeah. Um, We've got a rear end event. And look, you see the damages is far lower. The damage grow of that's 24, you know, the previous ones are 84. So it's still it. It's, it's a very serious event. You can see the risk, uh, risk of that. I'm going to play it up to you. There it is. Um, going into that full, um, play that again. So, and there's another, there's another crush is what, which is a little bit mild than that. Um, but those are those three crushes that that fleet, um, experienced in that, in that year. I, I can see them as an underwriter, but I've also got every other event summarized by driver for fleet risk or and driver risk or which, um, gives me the overall understanding of the fleet. And that's a final grain understanding that would have, in fact, that fleet now makes sense to me. I understand it. Right. You've had three claims. Okay. I, I can see them. Um, but I can see all the other risks. I just like to show you one other thing because, um, when you get a really big fleet. Um, let's have a look at that. This is a really large fleet and road facing videos. This week's having seven to 9,000 videos a day. Come in. Okay. Wow. They're really large fleet. Well, I'm just going to, I'm just going to show you. We'll leave it on that point. Um, of which, like, let's have a look at light to you's day. There are 229. Quite serious videos. It, uh, hidden in that 9,000 and 290 medium. So there's nearly like 600 events, which are kind of quite interesting. And then there's like close to 8,000, which are of, of, of lower severity. And in fact, some of them, thousands of those. actually, there really isn't any risk at all. The only way that you would know that is to watch 9,000 video clips for that day. Now, let's talk about that for a minute, because you know, these video clips, depending on the dash cam company, could be 10, 15, 30 seconds, something like that, but in that range, I think I've even seen some companies that do 5 second clips, but you're going to need teams and teams and teams of safety dash cam watches to get a handle around that. And it's going to take you hours to do, and you're going to be relying on humans. Now, kind of going back to kind of humans who are very, very bored watching videos that are engaged. I don't know, I, you know, let's suppose they're engaged, you know, let's suppose they're engaged, you know, given the benefit of the amount. If you look at something like risk from video and you get five, five humans in a room, I guarantee you that there will be some division of opinion. Some of them are going to say it like that, some are going to say it like that. Now, you can try to create house rules, but you're going to then, if you really want to get that right, you're going to have to put auditors on top of those reviews. So you're going to have to build this whole tiered structure of dash cam watchers, dash cam auditors, dash cam safety experts. Now, if that's 9,000 video clips a day, just from the road, I'm not talking about the cabinet, I'm talking about just from the road, you will need an enormous organization, which is going to cost you a lot of money. And the best you're going to do is broadly do that accurately, broadly. And we do it, not broadly accurately, we do it with complete accuracy in milliseconds, and you have to do anything. And that's what builds your driver's score. So it's not just about giving, well, it is about giving information to the insurance business, and you can see how an insurer coming to the fleet, I said you sounded the one that went out of business. If they wanted to understand, tell me about the last four months. Sure, it's not that I don't want to know there are 25 class eights, it's not that I don't want to know the cargo, it's not that I don't want to know the five years lost runs and stuff like that. But I also, I know a lot more about what's actually going on in the last quarter, the last six months with drivers now, whether there's been changes. And so I can really get a real read on the risk, like, you know, measuring the pulse of a patient. I can be fully up to date. But I go to the fleet, I'll also be able to think like, you know, again, I think I probably mentioned lost on this book, I'm not actually from the insurance sector, but, you know, if I wasn't underrated, I'd probably, I'd want to do two things. First of all, I want to measure risk very, very accurately and contemporaneously so that it was, my measurement was up to date. It wasn't, you know, somehow out of date for the reasons we described. I want to measure risk very accurately, but secondly, I would want to reduce the risk. So we give fleets a tool, the ultimate safety review, so they can reduce risk. So that's, that's, that's really what the product does. So it kind of, it leverages in existing infrastructure dash cams that are being put in place by, you know, fleets over the last year. We'll struggle to find a significant fleet in North America that hasn't invested in dash cam infrastructure, but you will also never find a fleet that's using those driver schools for insurance, because the schools don't do our correlate coins. So we solve that problem. We basically, we don't, we don't solve the dash cam solution. That's, that's what other companies do that, but we connect that down the, the video and the value of that video to the insurers and we also give the fleets a tool so that they can automate fleet safety. So even a fleet with 20,000 vehicles can have one safety officer versus whole teams of them. So that's really what the product is. I mean, we call it fleet vision. It, yeah, it, it makes intuitive tends to me as a former dispatcher as an underwriter and, it, it feels like the missing piece between dash cam and actually any information that we can actually do something with, right? So yes, awesome, awesome, that we got the dash cams. They're, they're good for, you know, post-accident review, but they're not good for underwriting, unless we have some sort of, or reduce schools that are not predictive of claims or losses, unfortunately. And that's, hence, the three and a half thousand experiments and not wanting to nerd out too much, but we use the kind of machine learning approach, AI protocol, N2M models, which are more powerful, but innately more complicated to get working. And so it's, it's really only when you use those kind of really state-of-the-art techniques that you can use for the extract value from this video. So again, it's a layer that sits on top. It's not a replacement for dash cams. This means to externally dash cams, but it's a layer that sits on top of the dash cam system and makes that information much more useful for the, for the, for the insurer, but also the safety manager for the fleet. Yeah, it turns that, that, just, barrage of data, we would, it could be thousands of videos a day into actually usable insights that we can use to prevent risk and to underwrite properly. 100%. It does one other thing. It nails ethanol in a way that is insane, okay? So, if you have a crash, sometimes it doesn't actually need to be claimed, okay? If your vehicle has a contact event with another road object, you know, it could be a sign, it could be a best drink, it could be a car, it could be a bike. It's going to tell you about that. So, it stops under reporting. It doesn't require serious crashes, serious claims. It nails ethanol, right? We're, we're running a large fleet that's putting in 50, 60,000 video clips a week and it's calling out 12 and saying those are the crashes. I mean, I showed you in the, the, the, the, the, the, the, the 25 class 8 fleet, you know, there's three crashes right at the top. It doesn't 100 videos. You don't need to watch any of them. If the system says you've crashed, you have crash and the crash doesn't necessarily mean a client, but it also estimates damage. So, you get an estimate in milliseconds from a contact event. That is amazing. So, the ethanol process is completely reinvented. It's totally dynamic. If you, unfortunately, have had it and you will, large fleet's world contact events, claim events, recently, frequently, given the, you know, volume of vehicles that go on the road. And you will know in milliseconds, I have had a claim event, or potentially a claim event, which means you can get to it immediately as a fleet. And the, it's interesting what, the largest fleet we're doing at the moment in the UK has got about 7,000 vehicles. And they, using a, a well known American dash cam system, which does actually also attempt to tell you if you've had a contact or crash event. And we were going through the amount of times that that was accurate. And it was basically about 30% to 40% of time. So, about 60% of the contact events will be missed. And anybody on dash cam will, knows that means you basically have to call off the driver, go into the dash cam system, try and find that segment, or that minute, or 10 minute, download the video, watch through it, try and pinpoint the point. Anybody that's working for your safety will, I think, empathize with that. We're on close to 100%. So, if you have a contact, that's it. And you don't need to, as a safety team, you don't need to work, watch anything else. Everything else will be done automatically for you. Obviously, if you've had a contact and a claim event, you want to watch it immediately, because you want to get at the event in real time. Maybe you can help out. Maybe you can assist with the resolution of the claim in real time, which is a big thing in commercial alternatives. And it, it really is a massive problem with the final mile fleets, where either they've got dash cam systems, and even though they've got teams of people who are doing the video groups, these are contact and positive claim events are missed all the time. And then they come back to the intro of three to six months later, as they puckerlate through the system. That's over. So, ethanol is solved. Full stop. I want to say that again, FNOL is so full stop. In short, vision is close to 100% accurate on F1 for FNOL. That means precision and recall. If you have a claim in milliseconds, you will know that you've had a claim or contact event and the estimated severity of that event. And that means as an operator, you can get to that immediately. So that's the kind of breakthrough for the industry. So we've seen ecstatic response to that. Yeah, it makes perfect sense to me. Absolutely perfect sense to me. I'm always curious of the implementation when it comes out. So we're a carrier or an MGA focused on commercial auto. We reach out, we see that we love it. What does the integration look like? In this case, correct me if I'm wrong. I don't think it has to integrate into my underwriting system or anything. Right? You know, I don't. You wish that you can integrate a system as much as you want. But what's it going to tell you? It's going to tell you the risk of the fleet and the risk by drive. Obviously, if you want it to integrate those numbers into some system and consume them, but you can just watch them. You can review them in a portal. So the information served up, we produce a front end for insurers, and sell it up for them in the format. There's quite a few things I haven't kind of shown, maybe for another episode, just showing the evolution of risk by driver, you know, by quarter, by fleet. You know, where that fleet is placing relative to its vehicle type, all the sorts of other things. Maybe for maybe maybe that's for another episode. But no, there's no real integration. You can use this. You get going immediately. What is pricing philosophy? This is our pricing philosophy. Yeah. And for the listeners, we're recording in early 2025. Things will change over time. Yeah, we price on the vehicle type on a monthly basis subscription. Super easy. Sorry, perfect. Yeah, pretty much that's how telematics has always been priced. It's always been a monthly subscription model. So some things, no matter what you want to do in an industry, you need to conform to some of the norms. So it's a monthly subscription. And that aligns with telematics more broadly as an industry. So it's priced the same way as every other telematics solution. Perfect. I'm always curious what's on the development pipeline? What are you excited for for the rest of the year? And what do you think is the future for this sort of solution? So I think if I was exclusively to talk to you about cloud processing and review, I'd be missing the risk. The best thing you can do if risk is to deal with it in real time. OK, like as it presents. And so the hash cam systems do have a lot to. But the problem has really been historically that the dash cams have been available. And we're built on quite legacy chipsets. Well, they weren't legacy when these systems were designed. OK, they were the chipsets that were correct for that moment. But machine learning, AI has moved really a lot in the last especially three or four years. So if you're designing a product through four years ago with a chip from three or four years ago, it's not really going to be very useful for machine learning of 25, 26 period. So what we've seen is we've seen the first generation of high power dash cams come out, hit the market from a few manufacturers. And I think there's more set to follow. But we've seen quite unusual things being done with those powerful chips, like things that we wouldn't do anyway. So I think what's the next thing for us is all of the models I'm showing you, they're all on the edge. They run at the edge on really very low levels of compute. Not like you can't run it on an eight-year-old chip because you just can't run this type of machine learning with something with essentially no compute. But we can run these models at the edge at very, very high frame rates on the current generation of cost-effective edge chips. And so there's three companies that I think about. There's Nvidia has got a chip called the Jetson-Orren. Remember, row up has got a CV 75 series, and you've got a wild clone got a chip called Dragonwing. And that's the kind of current generation of IoT edge chips. And we can run at very, very high frame rates all these models at the edge. So what does that mean? The system can tell you that you're going to crash before you crash to stop you crashing. It doesn't have to tell you afterwards. But it also means no false positives. The thing that really, really affects dash cams at the moment, especially when we talk about road understanding, which is complicated, is false positives. And we operate on a rules as a company, which I think is rooted in the way humans are. But even if I tell you something that's right, 60% to 70% of the time, that's wrong, 30% to 40% of the time, you'll basically ignore me. Yeah, absolutely. A comeback would be, hey, you shouldn't ignore me, because I'm more right than wrong. But it's just the way humans have been designed. If you want to engage with a human, you need to be right more than 90% of the time. Now, dash cams struggle with that. They struggle with false positives. And so the-- it's not that they don't get this thing right some of the time. They do. It's just they get it wrong quite a lot of the time as well. And so what we've done is we've built real time, fresh and risk understanding to the edge. That can be deployed on the newer generation of dash cams. And that represents a revolution in aftermarket ADS. So I think that's really exciting. As a business, that's something we've focused on. Relatively low power chips. Very high frame rates. Because we don't have very high frame rates in a lot of latency. And delivering to edge chipsets. So expect-- I think expect that whole real time risk reduction to really move forward, really meanfully. That's all prediction going forward. And to bring this benefit from the cloud to the edge. Fantastic. Fantastic, fantastic, yes. And the more that we can eventually move towards preventing it from happening, rather than dealing with it after it happens, is a win for everybody involved. I think so, yeah. I think so. And I think people will tell you that if you really, really want to understand everything about a drive, you need to run-- you need to run your eyes over essentially everything the drive is doing. And that means edge deployment all come out. All right. Mark, fantastic. Congratulations on launching. Thank you. Congratulations on hitting CES. That's awesome. I can't wait to see this being widely deployed. I'd love to have you back. You told me. Thank you very much, sir. Thanks so much. OK, sure. Bye-bye, Simon. Bye.

Podcast Summary

Key Points:

  1. Mark Miller of InsurVision.ai launched a new product, Fleet Vision, at CES 2025, which uses AI to analyze dash cam video data for real-time risk assessment.
  2. The system identifies crashes and contact events with near 100% accuracy, significantly improving detection over traditional dash cam systems that miss up to 60% of events.
  3. Fleet Vision provides insurers with detailed, actionable insights—including driver-specific risk scores and fleet-level performance—without requiring deep integration into existing underwriting platforms.
  4. The solution runs on edge devices with modern IoT chips (e.g., NVIDIA Jetson, Rockchip CV75, Dragonwing), enabling real-time risk prediction and crash prevention, not just post-incident analysis.
  5. By reducing false positives and offering instant event detection, the technology enables faster claims processing, better driver accountability, and proactive safety improvements.
  6. Pricing follows a simple monthly subscription model, aligning with telematics industry norms and making adoption accessible for MGAs and insurers.
  7. The company has conducted over 3,500 experiments to refine its AI models, which use advanced N2M machine learning to extract meaningful risk signals from video data.
  8. Future development focuses on full real-time risk prevention at the edge, shifting the industry from reactive to proactive fleet safety management.

Summary:

ai shares an update on his company’s groundbreaking Fleet Vision product, launched at CES 2025. The system leverages AI and machine learning to analyze dash cam video data from fleets in real time, providing insurers with precise, up-to-date risk insights. Unlike traditional dash cams that miss up to 60% of contact events, Fleet Vision achieves near 100% accuracy in detecting crashes and estimating damage—enabling instant, reliable FNOL (First Notice of Loss) processing.

The solution doesn’t require deep integration with underwriting systems, instead offering a user-friendly portal where insurers can review risk by driver, fleet, or quarter. It runs efficiently on modern edge chips like NVIDIA Jetson and Rockchip, allowing real-time risk prediction and crash prevention before incidents occur. This shift from reactive to proactive safety reduces false positives, improves driver accountability, and helps fleets reduce risk with minimal staffing.

Priced on a monthly subscription model, it aligns with existing telematics pricing norms and is designed for scalability across large commercial fleets. With over 3,500 experiments behind it, the technology offers a deeper, more accurate understanding of driver behavior and fleet performance—transforming how insurers assess and manage risk in the commercial auto space. The future focus is on expanding edge-based, real-time risk prevention, marking a significant leap in automotive safety innovation.

FAQs

Inturbation is a technology company that provides a video-based risk assessment system called Fleet Vision. It analyzes dash cam footage to offer insurers real-time, accurate insights into fleet risk, including driver behavior and crash likelihood, enabling better underwriting and risk management.

Fleet Vision extracts and analyzes road-facing dash cam videos to identify crashes, contact events, and driver risk patterns. It provides insurers with precise, up-to-date risk data—such as total exposure and driver-specific risk—allowing for more accurate and timely underwriting decisions.

No, Fleet Vision does not require integration with underwriting systems. Insurers can access risk data through a dedicated portal and review it directly. Integration is optional, though data can be consumed into existing systems if needed.

Fleet Vision is priced on a monthly subscription basis, similar to other telematics solutions. This model offers flexibility and aligns with industry standards for vehicle risk monitoring services.

Fleet Vision detects contact events—such as collisions with signs, bikes, or vehicles—using AI and machine learning. It provides instant alerts with estimated damage severity, reducing claim underreporting and enabling real-time response with up to 99% accuracy.

The company is advancing real-time risk detection by deploying AI models directly on edge devices in dash cams. This allows preventive risk reduction—such as predicting and stopping crashes—before they occur, reducing false positives and improving safety in real time.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.