What's Up with Tech?
Tech Transformation with Evan Kirstel: A podcast exploring the latest trends and innovations in the tech industry, and how businesses can leverage them for growth, diving into the world of B2B, discussing strategies, trends, and sharing insights from industry leaders!
With over three decades in telecom and IT, I've mastered the art of transforming social media into a dynamic platform for audience engagement, community building, and establishing thought leadership. My approach isn't about personal brand promotion but about delivering educational and informative content to cultivate a sustainable, long-term business presence. I am the leading content creator in areas like Enterprise AI, UCaaS, CPaaS, CCaaS, Cloud, Telecom, 5G and more!
What's Up with Tech?
Sensors, AI, and Incentives Are Making Our Roads Safer
Interested in being a guest? Email us at admin@evankirstel.com
The technology in your pocket—sensors in your everyday smartphone—is quietly revolutionizing road safety in ways most of us never realize. In this eye-opening conversation, MIT professor Hari Balakrishnan shares the remarkable journey of Cambridge Mobile Telematics (CMT) from university research project to global safety leader.
What began as a simple idea to use mobile device sensors to measure driving quality has evolved into technology that's prevented over 90,000 crashes and saved 50,000 lives worldwide. Balakrishnan reveals how CMT's DriveWell platform fuses data from phones, windshield-mounted tags, connected vehicles, and dash cams to create a comprehensive picture of driving behavior that's transforming the insurance industry.
The implications are profound. For the first time in insurance history, companies can price policies based on how people actually drive rather than demographic assumptions. Young drivers no longer automatically face punishing premiums—they can earn discounts through demonstrably safe driving. This creates a virtuous cycle where better driving leads to lower costs, fewer crashes, and ultimately, lives saved.
Perhaps most fascinating is CMT's breakthrough in real-time crash detection—something once deemed impossible using only smartphone sensors. Unlike solutions that focus only on catastrophic accidents, CMT's technology works across all crash severities, providing immediate assistance when needed most.
Looking toward the future, Balakrishnan shares his vision of data-driven road infrastructure improvements and the gradual integration of autonomous vehicles. All while maintaining a refreshingly human-centered approach to privacy: "Don't do anything with the data that you wouldn't want done with your own."
If you're interested in how technology can create safer roads while giving people more control over their driving costs and safety, this episode offers an inspiring glimpse into that future—one that's already arriving on roads worldwide.
Discover how technology is reshaping our lives and livelihoods.
Listen on: Apple Podcasts Spotify
More at https://linktr.ee/EvanKirstel
Hey everyone, fascinating topic today about making the world's roads and drivers safer with Cambridge Mobile Telematics. Hari, how are you I'm doing great? Evan, how are you I'm doing well? Wonderful mission. Thanks for joining to talk about it. Before that, maybe talk about your bio and background the journey from MIT research to really global impact at CMT.
Speaker 2:Yeah, happy to do that. I'm a computer scientist by profession, professor of computer science and AI at MIT. I've been here a little over 25 years. About 20 years ago, with my colleague, sam Madden, we started a research project called Cartel. This was back before iPhones and Androids. I had this idea that we could use sensors that were becoming more popular on mobile devices to measure driving quality and understand transportation, understand why our road infrastructure has problems, understand why traffic problems arose and, over time, also why crashes were happening, and that research project was hugely successful in the academic context.
Speaker 2:We won a number of awards and the press started writing about some of the work we were doing.
Speaker 2:One of them projects was the Pothole Patrol, where, in 2007, 2008, we instrumented our devices with accelerometers and GPS to produce every week a ranking of the worst potholes in the Boston area.
Speaker 2:This got written up in the Boston Globe and the Wall Street Journal, and I had this idea that maybe we could use sensing for social good, for solving societal problems in transportation. I then met my co-founder, bill Powers, who was CMT's CEO in late 2009. We started the company shortly thereafter, in 2010., and today we are the world's largest telematics provider, building mobile sensing and artificial intelligence technologies to measure how people drive, to provide incentives for better driving working by our partners in insurance and ride shares and the commercial space and then to also detect crashes in real time, when they happen, regardless of severity and regardless of speed, and provide assistance in the form of roadside assistance or emergency or towing and that's really been the journey or emergency or towing, and that's really been the journey. We're currently about 475 employees headquartered in Cambridge, massachusetts, but really a global company with offices in many countries, and we serve our users millions and millions of users in about 25 countries around the world.
Speaker 1:Amazing, and that journey sounds so straightforward and simple and seamless.
Speaker 2:It was, it was, it was.
Speaker 1:From the early days doing mobile sensing experiments it sounds like to truly global platform, I guess. Billions of trips across 20 or 30 countries that's extraordinary.
Speaker 2:Thank you so much. It was an overnight success that took over 10 years. So how did we scale? I think, first and foremost, I don't think there's a single magic bullet secret, but first and foremost, I would say that we've always thought of ourselves as trying to understand deeply the problems that our users and our customers face and to try to solve those problems with whatever technology and products we could bring. So we've avoided the mindset of pushing our products or our technology onto people. We really are trying to fall in love with the problem and then use our know-how and our ingenuity and our hard work working again really in concert with our partners to solve their problems. So I would say that we try to help our customers grow and therefore grow ourselves. That's probably been a key aspect of what we've done Very collaborative, it's. One of our key values is to be highly collaborative, both internally and with our partners, and try to focus on simple solutions, try to come up with the simplest way to tackle problems, because these problems are extremely complicated.
Speaker 1:Complicated indeed. And you not only fuse IoT data from phones, but there's tags, there's vehicles, dash cams, all other kinds of feeds. How do you orchestrate all of these you know diverse data streams into a single picture of safety?
Speaker 2:That's a great question. It has been an evolution of how we do that. When we started, we were a single data source. We started with sensors on phones. You know, every iPhone, android phone out there for the last 15 years has come with accelerometers and GPS and gyroscopes and so on, and that's how we started. Soon, we invented a small 5 centimeter by 5 centimeter device called the Tag. That's an option. It works together with a phone. We've shipped over 40 million Tags around the world. We've shipped over 40 million tags around the world.
Speaker 2:It's one of the most widely used aftermarket automotive products out there, purely for the purpose of understanding, driving and providing crash assistance and fulfilling our mission of safety. And then you know, we process data from connected vehicles, from vehicular dash cams and so on, and we've created this platform. We we call it the DriveWell Fusion platform. Drivewell is our suite of products and Fusion is the platform on which it runs. Drivewell Fusion brings together data from a disparate, wide variety of sources and harmonizes it, and the harmonization means that, regardless of the type of data source, we provide actionable insights and output data that allows for business applications, such as for pricing your insurance or understanding your safety score or crash and claims or behavior change to provide incentives for better driving.
Speaker 2:Those business applications don't have to depend on the vagaries and the peculiarities and the differences of the data. So it's not just that we make the data schema look the same, the data meaning is also the same, and to do that requires a lot of advances, both in understanding the physics of movement patterns, the dynamics of vehicular movement, as well as the dynamics of how people, for example, use their phones within the vehicle while they're distracted. There's patterns of movement inferred from the accelerometer of the gyroscope of the phone, but it's really a physics part. There's a signal processing part and, of course, there's a more modern machine learning and artificial intelligence component. So if we bring together physics, signal processing and machine learning, we can do a lot better than with any one of those methodologies alone. And that's really been the secret of our know-how as a team that is equally adept at solving difficult physics-oriented problems or signal processing problems as with machine learning Amazing.
Speaker 1:And, unlike our friends at Big Tech, you really advocate for a privacy-first kind of consumer choice-led model. What does that mean in terms of balancing all that good, great data you have with user trust and getting all those insights on user behavior?
Speaker 2:Yeah, I won't speak for Big Tech, but what I will say for us is there's a very simple value principle Don't do to or with the data what you don't want to do with your own data. Imagine yourself you are a user. If you're an engineer, a product person, an employee of the company, don't do anything with the data that you don't want to have happen to your own data. Everything stems from that. We have a road safety board. We've had a very strong industry privacy board in the past, led by some of the leading thinkers at the intersection of technology, privacy and law, and we've crafted a set of rules that are simple to understand, and they all boil down to exactly what I said Don't do anything that you don't want to have happen to your own data.
Speaker 1:Well said, and you work with some very well-known, successful insurance companies. I won't name them all. I certainly probably use one or one of them. But how do those partnerships work? What do they look like and what have they taught you about behavior-driven technology?
Speaker 2:Yeah, we are very fortunate to work with some of the largest and most successful iconic, long-running companies. These companies, you know, in tech we have companies that are, you know, 10 years old, 20 years old, maybe 50 years old, but in insurance we have companies that are 100 years old and it is just. I can't even imagine what it must take to have a company that lasts that long and thrives. So we've been very fortunate to work with them and learn from them. The model is actually not that complicated. It's based on sharing a vision for how we want to help them succeed, and that is to come up with a more predictive and more equitable form of pricing insurance. And with telematics, for the first time in the history of insurance, you have a way to set prices based on how people drive and therefore provide more control to our users, because if you know how your insurance price is being set based on your driving behavior, you can actually do something about it. You can stop the phone while driving, you can stop inattentive driving, you can reduce the amount of excessive speeding, you can be more attentive to. For example, take a young driver 23 years old.
Speaker 2:My daughters are in their early 20s.
Speaker 2:Traditionally, they would have had a terrible insurance price because, statistically, young drivers have a higher risk, but not all young drivers have a higher risk and, moreover, when you give them the feedback and you give them the incentives, they can become better drivers.
Speaker 2:And right now in my family, my two daughters are the best drivers and that's because they're part of a behavioral program that gives them rewards for better driving. So all of a sudden, that high risk has gone down. It's good for them, it's good for a society with fewer crashes, it's good for the insurance company because they now have, you know, they can collect some premium, they give them a break on insurance but at the same time they're going to hopefully have fewer claims. And certainly it's certainly good for CMT as well, because we achieve our mission, we drive up our growth and therefore it's a win-win. And the work is very partnership oriented. We understand the business goals of our insurance partners or ride-share partners, for example Uber, and we try to solve those problems with them for the end user. So it's really trying to put the end consumer first and providing the right type of incentive structure to make them want to use it and then benefit from it.
Speaker 1:Wow, talk about a win-win-win type of incentive structure to make them want to use it and then benefit from it. Wow, talk about a win-win-win. Let's talk a little bit more about the technology involved. When was the sort of aha moment when you realized mobile devices could become sort of precision safety tools?
Speaker 2:Yeah, that's a long time ago. I wrote a paper with my colleague Sam Madden and our students back in. We were working on this project in 2004, 2005, 2006, and the paper was published in 2006. It's one of the more cited papers in computer science. It's called Cartel, a Mobile Sensing System, and there's a sentence in the introduction that says that soon these will be available on smartphones and mobile devices. And again just going back to that time, it was predates, the iPhone and the Android ecosystem. It was Nokia phones which did have sensors and I remember the Motorola Razr phone that flipped. I hear it's coming back now, but those were the phones that we were looking at.
Speaker 2:But to me I don't want to say it seemed obvious, but it seemed to me inevitable that right from the late 90s, when again predating Wi-Fi, it seemed inevitable to me that, given technology trends, the miniaturization of computing was happening at big scale. And I got invited to some workshops very early as in my graduate school career, at the end of my graduate school career, very early in my own career, where they were talking about how sensors, mems technologies, were getting smaller and smaller and more economical and wireless communication was becoming more widespread and smaller and smaller. So when everything becomes smaller and smaller and faster, it seemed inevitable to me that this would happen, that we would be able to build devices and the only question was whether we could make them battery efficient. Because they were small, they'd be powered by batteries and I wouldn't say the only question there was the question of battery efficiency, there was a question of cost and there was a question of accuracy whether these consumer devices they're super low cost. And you have to remember, back 25 years ago, sensors were used in a lot of applications like avionics, the military and industrial process, and they were really good sensors, but they were expensive sensors. People would pay $5,000, $10,000, not $5,000 or $10,000, but $5,000 or $10,000 for these expensive sensors. And so the question was whether we could get them to be accurate enough at low cost and energy efficient.
Speaker 2:And our research and development work at both at MIT and then at CMT, as well as the entire industry, really worked very hard. And, of course, the watershed moment came when really the iPhones and then Android. Around that time that ecosystem exploded and when we started it was not possible to have background sensing applications run on iPhones. So we built our first prototype and research. We did research on an old jailbroken iPhone where we got root access. We bought an iPhone.
Speaker 2:We just wanted to show that it could work feasibility-wise and we could, which meant that if Apple wanted, they could support it, and we placed a bet that, yeah, you know what they probably will, because there are so many useful things you could do with it outside of road safety. You could do health and wellness. You could do things like notifying people of you know things that they need to do, and so on and so forth. And then we did it with Android in the first in 2012, when we launched our first mobile probably the world's first mobile usage based insurance pilot program with a leading insurer. It was only available on Android. It was only in 2013 that Apple iOS opened up this capability to developers, but we were right there when that happened.
Speaker 2:And then when we developed the tag, again based on a lot of feedback from one of the early strategic partner discovery insurer in South Africa, it was really a product that almost looked impossible but ended up being possible. And this was the first completely battery-operated small device that you could just stick to the windshield and would have four years of battery life. At the time, a little bit better, even better now and the cost has come down. And again, the idea of taking this type of telematic sensing technology that was previously available in industrial applications, that was previously available only to large fleets, which were paying a lot of money, and making it oriented toward consumer products so consumers could use it easily and at really low cost. That really was one of the key innovations. And, of course, you don't end up with the most accurate sensing data set, and that's where the technology comes in. That's where the AI and the physics and the machine learning and the signal processing come in, which is we're able to handle all of the errors in the raw data to produce really good, accurate outputs.
Speaker 1:Amazing. You also solved something that was said to be difficult, if not impossible, early on identifying crashes in real time from mobile sensors. Of course, recently we've seen Apple baking that into the iPhone and the watch, but what enabled that breakthrough on that end?
Speaker 2:That's a difficult question. It's actually a problem. It still remains a very difficult problem to achieve extremely high precision, which is to say no false positives, and high recall, which is capturing every crash, and there's a trade-off. So, for example, our technologies achieve we are oriented toward high recall. Our customers and our users want to catch every possible crash as possible and we focus on that. Apple has it in the iPhone, on the newer iPhones and the newer watches, and that's focused on getting the very highest impact, life-threatening crashes, whereas ours is across the spectrum, including roadside assistance and claims using insurance claims. So it's just picking different points in the spectrum.
Speaker 2:I'll be honest, I thought back in 2014 that this could be really difficult, if not impossible, to solve on a pure mobile phone device. So one of the motivations for our tag was to develop a windshield mounted device that could do it, and actually what's interesting is that because that device is windshield mounted and we have millions and millions of them every day in the field it's a little easier to solve the problem on a windshield mounted device because it's not being moved around on the body of the person and it's not in your pocket where it could get damped. The signal could get damped, or somebody could put it in a handbag and leave it in the backseat and maybe it doesn't feel the shock the same way. But we have this almost unfair advantage, which is we have a windshield mounted tag on 60% of our users, which means and they collect both the phone data and the tag data, which has allowed us to really build models that work with any data source, because lots of our users are providing us two data sources that work with any data source, because lots of our users are providing us two data sources. Second, we have deep partnerships where we're getting ground truth on what actually happened and we work with our partners to provide them even better results that are tuned to their demographic.
Speaker 2:Countries are different. The type of driving is different in different countries, so we're able to then adapt our models. We have a base set of models and then we're able to then adapt our models. We have a base set of models and then we're able to adapt them to particular kind of conditions, and all of that has allowed us to succeed. But I'll say this we believe we are best in class. We have the most experience and the most accurate outcomes, but we're still working really hard on further improvements and there have been breakthroughs in AI. You know we now have models that we use that are quite interesting in that they have some inspiration from take some inspiration from the type of models used in chat, gpt for natural language. So I think advances in AI continue to appear and we're able to benefit from some of those advances in the research literature as well.
Speaker 1:Fantastic and the impact is clear. I mean, I see you're credited with helping prevent over 40,000 road injuries and fatalities. That must be very gratifying, but at the same time, big picture, at least here in the US, we have a pretty poor driving safety record compared to our peer rich nations, let's say really concerning fatality rates. You know, rich nations, let's say really concerning fatality rates. What's your takeaway from that? How can we or you impact even more change for good?
Speaker 2:Yeah, that's a great question. I think our latest numbers are that our analysis shows we've averted over 90,000 crashes and 50,000 injuries and fatalities, and I think these are big numbers, but they're not anywhere near what we need to do at national or global scale, and we're working hard with our partners on this, Okay, so what do we need to do and why are they problematic? So I'll give you some bad news and then some good news. So the bad news, in some ways, is that since, especially since 2017, vehicles have come laden with more and more technologies. A lot of it is around safety and yet, if you look at the data, for several years, crash rates have not kept up in terms of, they've not come down at the same proportion.
Speaker 2:I call that the vehicle paradox, and one of the reasons for that is that the technology actually is quite complicated. For example, you have a blind spot monitor, which is good. You might have one of those things that keeps you in your lane, but sometimes what happens is it pulls you. You're trying to move lanes and it pulls you back and you're wrestling with the car to stay in the lane. So these technologies have more complicated human machine interfaces, much more so than they used to be, and cars have become quite different from each other in terms of their capability, and sometimes people who drive two cars end up getting confused. So there are all these issues that I think are one of the reasons why crash rates haven't come down as much.
Speaker 2:And then we have phone distraction, which is a huge factor in terms of causing people to lose attention or not pay attention and get into crashes. The US continues to have the highest rate of phone distraction per mile or per hour of driving compared to most other countries. It's a place where we're number one when we shouldn't be, at least among the countries we've studied. But on the positive side, we've seen about a 9% drop in the amount of distracted driving per hour for users in our program over the last year, and we think that's because these programs are now scaling. There's lots of incentives in place. There's punishments in place where people's premiums might go up if they get into trouble because of crashes, and then there are laws against distracted driving that have you know that have come about in many states. I believe over 30 states now have laws against phone distracted driving, and that and CMT's broader analytics studies not, you know, looking at individual users. But just looking at overall rates suggest that at least in the several months after laws pass they do help and of course after that it depends on the rate of enforcement. So I think all of these things, there is some promise that the distracted driving rate is coming down.
Speaker 2:I do think that these programs, when we see examples of successes some of our partners, for example, Discover in South Africa sees a 26% lower crash rate compared to the rest of the population for people engaged with the behavioral program. In Japan, our customer IOE sees an 18% lower crash rate and even in the US there are successful examples amongst our insurance partners where the majority of users in the program see discounts. In the program see discounts, the majority of users see discounts. The majority of users actually have better behavior in terms of their driving quality and lower crash rates. But you know we continue to do the work because the overall penetration in the markets for users who use this on a daily basis is still in the high single-digit percentages. So we have a long way to go. We've done well but we have a long way to go.
Speaker 1:Long way to go, indeed. And what does the next decade of road safety look like? So many emerging technologies, smart glasses, vehicle-to-vehicle communication. What else is on the radar that will make our road safer?
Speaker 2:Yeah, there's a lot, I think, going on. I think, first and foremost, I think when we scale up the telematics-based programs that CMT works on, I think we will see dramatic changes in human driving behavior. Despite the complexity of vehicles, I think these programs work. I think we'll continue to see a reduction in distracted driving. I really believe that. I think there's a lot of data available now, a lot of incentives to prevent it and penalties as well, so I think that'll make a big difference. Number two I mean I think the road to autonomy will continue. Robot taxis are coming soon. I think that their penetration into the consumer driving market will take more time and it'll be a long, long haul because the average age of a vehicle in the US on the road is about over 12 years, so people are unlikely to just replace their car to get an autonomous vehicle. But it'll happen over the next 15 or so years, maybe 15 to 20 years, and then I think that will really be dramatic and in that world I don't think crashes are going to be eliminated. They're going to be lower but different, and anyone doing insurance or safety in that world will be fundamentally based on telematics, because no longer does it matter you know the color of the skin, or I don't think they use that in pricing. But the type of car you drive, the color of your car or the age or demographic, the place you live, may matter a little bit because of crash rates, but whether somebody is 20 years old or 50 years old won't matter so much. But the way in which you rate drivers will be fundamentally based on telematics data the way in which you rate the robotic drivers, and that has to do with evaluating the quality of software, the quality in which you rate the robotic drivers, and that has to do with evaluating the quality of software, the quality of the sensors and the quality of the AI. So we think CMT is well positioned 15 years from now.
Speaker 2:In the interim time, one of the areas where we'll see dramatic improvements is road infrastructure. Data-driven decision-making in road infrastructure is an area that we're pioneering and working on. It's part of a product we have called Street Vision, which uses aggregate, anonymized data to inform cities, municipalities and states about how to improve road infrastructure Placement of stop signs, placement of speed bumps, bus stops, highway speed limits, suburban roads, urban roads. How can you change infrastructure, both in the short term to lower crashes, to determine the effect of an intervention before it happens. It's very difficult to put a stop sign, go about measuring it and then deciding if it was a good idea or not. What if we use data and AI to make the determination before? That's what strict vision is up to, and then, as we head on the road to autonomy, we think infrastructure has to change that.
Speaker 2:We think that there's going to be infrastructural changes needed to allow for the coexistence of autonomous and semi-autonomous and manual vehicles, and not to mention just not just the vehicles autonomous and semi-autonomous and manual vehicles, and not to mention just not just the vehicles.
Speaker 2:We share roads with bicyclists and motorcyclists and, you know, pedestrians and runners and construction workers. Everybody has to be safe. This is important because, you know, 50% of serious injuries on roads and fatalities happen to people outside the vehicle. Serious injuries on roads and fatalities happen to people outside the vehicle. 20% of such injuries and fatalities in the US are to people outside the vehicle. So they have to be part of the revolution, part of the technological evolution, and I could imagine a future where we have clothing that we wear with sensors on them that keep us safe on the road, and I think that that's the future we're headed together toward. You know things like V2V, v2i. I think these may be enabling technologies, but they're very low-level communication technologies. I like to think top-down of the outcomes we want to achieve, and the outcomes we want to achieve are data-driven, ai-driven and behavioral-driven.
Speaker 1:Wonderful. Well, you're in the lovely city of Cambridge, Massachusetts. How's cycling in and around Cambridge these days Getting better, or is there yet work to be done? As you, and your team come into the office.
Speaker 2:Much better than it used to be. I ride my bike almost every day to work, and Cambridge has done a good job. They've tried a number of good experiments. For example, they have one where they put the bike lane near the curb and they have parking in the middle and then they have cars on the other side. That has two benefits Cyclists don't get doored as often and second, you have a shield between the bike lane and the traffic. They've tried a number of experiments, but, as with anything, anything it's a work in progress because, uh, you know, like with any societal problem, there's people who like bicyclists, people who don't, there are people who like drivers, cars, they don't. But ultimately we all have to coexist. There's only a certain number of kilometers or miles of roads and we have to share it. So, um, you know, boston's known for some uh, let's just say aggressive driving.
Speaker 1:Well, we're number one Of the worst drivers, that is.
Speaker 2:I don't believe we're the worst drivers. We have data showing where the worst drivers are Okay.
Speaker 1:We'll keep that data top secret, but thanks so much for joining us Really eye-opening discussion. Appreciate the mission and spending time here. Thank you, evan, thanks so much and thanks everyone for listening, watching, sharing this episode and check out our new TV show, TechImpact TV, now on Fox, Business and Bloomberg. Thanks so much. Thanks, Ari, Thank you.