Summary

This TU Automotive webinar, sponsored by Arity, delves into the critical issue of distracted driving and explores potential solutions. Hosted by Steve Bell, the session features insights from Katie DeGraaf, who oversees product management for mobility intelligence at Arity, and Carl Schmidt, a senior data scientist specializing in telematics data synthesis and insurance risk modeling. The discussion highlights the evolution of driving environments, the impact of sleep deprivation, and the increasing complexity of vehicle cockpits, all contributing to driver distraction. The speakers emphasize the role of telematics and mobile technology in detecting and mitigating distracted driving behaviors. They present data showing the correlation between smartphone use and increased accident risk, and discuss how insurers can leverage this information to reduce losses. The session also covers the psychological aspects of phone addiction, the effectiveness of various engagement tactics to change driver behavior, and the importance of a comprehensive approach to addressing distracted driving. The webinar concludes with a Q&A session, addressing topics such as the impact of hands-free phone use, the role of connected cars, and strategies for reducing insurance claims through better driver engagement. 

Transcript 

Steve Bell: 

Thank you very much, Becky, and welcome everybody to this TU Automotive webinar, sponsored by Arity on The Cost of Distracted Driving and What Could be Done to Address the Problem. This is a very real issue and to provide insight into the challenges, I am joined by two very talented individuals—Katie DeGraff, who oversees product management for Mobility Intelligence at Arity, including the development and recommendations of new techniques and processes to build innovative solutions for Arity’s insurance companies. Katie spent more than a decade with Willis Towers Watson, where she played a key role in launching its telematics rating solution. She has a deep passion for telematics and has spoken on the subject across the globe and also serves on the National Association of Mutual Insurance Companies’ Personal Lines Competence Planning Committee. So, welcome, Katie. I’m also joined by Carl Schmidt, who is Senior Data Scientist managing the advanced sensor insights team at Arity and specializing in telematics data synthesis and insurance risk modeling. 

His Arity colleagues he works with are working towards solving some of the toughest challenges of today’s mobile data telematics, including driver and passenger characterization, collision detection, and distracted driving detection. Carl spent six years in the energy sector utilizing predictive analytics and probabilistic modeling for the nuclear power system applications and has an MS in mechanical engineering from MIT. So, welcome to both of you. Today, our agenda is a short introduction for myself and then we’ll go on to two primary areas, distracted driving and detecting behaviors and nudging drivers, and we will have time for Q and A at the back end that I encourage you all to submit your questions in the Q and A box that Becky referred to earlier on. We will also have a couple of poll questions, which enable us to interact with you as the audience and get a feel of how you are reacting and also how you perceive the subject. 

So, without further ado, let me talk about distraction. Having heard some of the presentation and seeing some of the insights already, I wanted to try and put this into context. Now, a lot of what we’re going to hear is the driving, and the technology, and human behavior, but obviously there are other elements that impact distraction, including sleep, and, obviously, context. Driving itself has rapidly evolved. It’s no longer just open road, simple controls, and a radio. The complex cockpits of today have vast amounts of information and sensor data being provided into screens, touch screens, cell phones, multiple controls—in many cases, multiple screens providing the traffic info, congestion, and so forth. The problem is that the roads themselves have also become more congested. There is less time to get to where you’re going, although it takes longer. So, there’s a lot of distraction within the cockpit and on the roads themselves. 

Add to this the fact that sleep is a major issue for everybody. From 1942 to 2013, the average hours slept in America, dropped from 7.9 to 6.5, and it’s not just in the U.S. In a recent study, half the world is getting less sleep than they need and 80% are using weekends to make up the lost sleep. But, of course, one of the problems is that one of the most high-risk drivers is always teenagers. And, if you look at the statistics from Scientific America, 20% of teenagers are getting less than five hours sleep while the average is only getting 6.5. So, the least experienced, they’re getting the less sleep, and they’re also the most vulnerable for distraction from devices, as you’ll hear later. Obviously, the industry itself is trying to address this and, in fact, Volvo is putting cameras into cars to detect distraction, sleepiness, and drunkenness. 

The system will be able to slow the car, call the emergency services, and fully pull over if the driver is inattentive. But, nonetheless, it’s still a dangerous aspect that needs to be addressed. The other thing that needs to be looked at is that distracted driving and rideshare. We’ve all been and used these Uber, Lyft, and other rideshare facilities, and the drivers are constantly looking to find the location, find the fastest route, taking messages in terms of where their next gig is, and at the same time engaging with customers, and that’s all going on at the same time as they’re trying to navigate traffic and also manage the controls of the vehicle. So, the distraction factor for rideshare is actually increasing and one of the questions, obviously, becomes what does that mean from an insurance perspective? The other issue, I think that’s worth bearing in mind, is that distraction and the road towards the autonomous vehicle, the semi-autonomous vehicle, is part of that pathway, and the requirement is that drivers should be supervising, if not hands on the wheel, to the events and supervising the automation of the vehicle. 

But, as recent crashes suggest, remaining alert is not enough to take control, and, in fact, the Journal of Safety Research identified in some research they were doing that drivers using automated systems responded worse than those manually driving, in terms of reaction time, lane departure, maximum steering-wheel-angle-induced-in-lane departure events. These results also found that non-driving tasks further impaired the driver. And so, in an automated driving condition, driver’s response to safety critical events were slower, especially when engaged in non-driving tasks, which means that the risks are increasing on the road, not decreasing. So, with that, let me set that context and ask the first poll question. Does your company currently have an initiative focused on distracted driving? The answers are yes, no, or implementing in the next six months. And, at that point, I’d like to invite Katie to join me who’s going to be the first speaker. So, Katie, what do you anticipate the response to this question will be? 

Katie DeGraaf: 

Hey Steve, so, if this poll follows what we’ve seen in other aggregate industry research polls, we tend to find, especially among the insurer pool, that even digital and mobile strategy is not as prevalent as in other financial and other industries. So, let’s see what comes back here. I think recently we’re hearing more of a priority on digital and mobile first, followed by some things on distraction using technology. Most insurers do, though, have some sort of non-tech, sort of offline strategies to engage customers on the behavior of distracted driving. 

Steve Bell: 

Well, with this result we had, yes, 58% of people had some form of initiative, 29% no, and 12% implementing in the next six months. So, that’s probably consistent with what you’re anticipating, right? 

Katie DeGraaf: 

Yeah, that’s really great. And, like I said, I think most of these are non-tech initiatives, so using things like social campaigns and education in schools, which is great and very effective, we do think this partners well then with leveraging technology to get that information back to consumers in a meaningful format, and that’s what we’ll be focusing on today. 

Steve Bell: 

I’ll let you take it away. Thanks very much, Katie. 

Katie DeGraaf: 

Super. Thank you, Steve. Quick introduction of Arity before I hop into our content. We are a company of 400 employees across engineering, data science, user experience, behavioral science, and product. And we’re focused on understanding mobility risk. With over a decade of telematics data paired with actual claims and insurance policy characteristics, which includes the largest mobile database in the world, we partner with insurers and shared mobility companies to make transportation safer, smarter, and more useful for everyone. So, we’re pleased to be speaking to this audience about a topic that’s of highest importance to this entire company. We are located in the heart of Chicago, in the Merchandise Mart, and we have an open, dynamic and lively office. And you won’t be in this office for more than a few minutes without hearing references to distracted driving. You’ll hear our mobile devs seeking to optimize the data collection. 

You’ll hear our data scientists and actuaries who are seeking to understand the true behavior and the impact to insurance losses. You’ll also hear our UX and behavioral scientists continually experimenting and iterating on how to message and engage drivers to eliminate this dangerous behavior. And, as Steve mentioned, I’ve been in this space for about a decade and I’ve had the great pleasure to work with many in this industry to deploy telematics programs and we’ve worked really hard through many sleepless nights and iterations to bring solutions to the market that are appealing to consumers and also solve business problems. And since the beginning, we’ve all held out hope that we could use telematics as a platform to better understand our drivers and help them become safer. In my perspective, our progress has slowed of late in an effort to ease some of our operational deployments and hedge our bets against the future of insurance, we’ve been pretty conservative and slow in our enhancements on engagement and analytics, but this time is definitely our greatest opportunity. Mobile technology, consumer awareness of distracted driving, and the practices of UX and behavioral science have all converged during a time when our roads are more dangerous than ever and a primary driver of that distracted driving is through smartphones. And Steve, you also mentioned sleep and I’m glad you mentioned that. That’s actually a pretty big booth that’s swept through our offices as well. Most of us have read why we sleep or sleep smarter and we frequently talk about the number of hours of sleep we got last night. So, we certainly are looking at sleep as well, but we do believe the smartphone, it’s a primary contributor of distraction today, but it’s also the relevant tool to engage and motivate behavior. So, that’s what we’re going to focus on. 

So, during this session, I’m going to start us off with an overview of distracted driving, including the impact to drivers in our industry. I’ll highlight some findings from a recent consumer survey and touch on why we continually reach for our phones even while driving. Then, I’m going to turn it over to one of our lead data scientists, Carl, to talk through how sensors on the phone can be used to identify distracted driving and some key principles for using that data. Then, I’ll come back and close this out by sharing how in-app experiences can be used to motivate a change in behavior. What we hope you take away from this session is that the mobile phone is a relevant and meaningful tool that can be used by insurers to identify, engage, and contribute to ending distracted driving. And, as I tell myself before every workout, it’s not easy, but it’s worth it. For sake of an overview, why is this a problem? 

Well, an observational study going all the way back to 2006, found that looking away from the road for just two seconds doubled the likelihood of an accident. So, try it now. I’ll wait, look away from the screen that you’re likely multitasking on, and come back. That small behavior doubles the likelihood of a crash. Interesting that this study was done way back in 2006. This was one year before the release of the first iPhone, and there honestly wasn’t much pulling our attention away from the road for two seconds, but today it happens continuously, and, in fact, sending or reading a text message takes about five seconds. And it’s hard to tell exactly how many crashes are caused by distracted driving because it’s often not reported or admitted. But according to NHTSA, over 3,000 fatalities occurred in 2017 related to distracted driving, and our friends at the NSC, the National Safety Council, estimate that 27% of all accidents today are due to phone use. 

In our own survey of over 2,000 drivers conducted earlier this year, we found that nearly one out of four reported being in an accident caused by a distracted driver.

We also keep a pulse on the consumer attitudes on distracted driving. In our own survey of over 2,000 drivers conducted earlier this year, we found that nearly one out of four reported being in an accident caused by a distracted driver and 61% of young drivers admitted to engaging in phone use beyond phone calls. So, that includes texting and watching videos while driving. And, we also found that there’s certainly a widely held belief that this is everyone’s problem, but maybe not their own. While 98% report seeing others distracted, less than half of the total respondents admit to doing it themselves, only 42%. And, we know the laws are having minimal impact with just 4% of our survey respondents having been issued a ticket for this behavior. So, from this we can conclude that the problem is pervasive. We cannot rely on the enactment of laws to fix it, and there’s a gap in self-awareness. 

So, why are we so prone to distraction, particularly when it comes to our phones? It all comes back to our good friend, dopamine. Back in high school biology class, you probably learned that dopamine is a chemical that’s released in our brains when we’re happy, or when we experience pleasure, and it’s generally related to an instinctual need. For me, this happens every time I eat broccoli. That’s right, I’m not joking. I love broccoli. What that means, though, is that dopamine is playing a starring role in motivation and habit formation. So, it’s not that we just don’t have the willpower to stay off of our phones. These behaviors are actually dopamine fueled and hardwired into our human need for social interaction. 

I want you to try to think back to a time when you first got your first smartphone and you used it to connect with someone. It was probably pretty exciting and dopamine was definitely involved. With each dopamine-inducing event, like you get a text message, your brain starts to associate your phone with dopamine. And since your brain naturally craves dopamine, it starts to crave your phone or really the social interaction that comes from your phone. And, today, there are even more ways our phone initiates a release of dopamine, like Instagram likes and Snapchats, and the more it happens, the stronger that association gets. So, with each text, with each phone call, with each notification, it’s increasing the intensity at which our neurons both crave the interaction and respond to it. 

Unfortunately, it’s not just the release of dopamine that contributes to our addiction, but also the frequency and rate at which we’re having these experiences. Our phones and many of the apps we use leverage a reward pattern that’s optimized to keep you engaged as much as possible. For all my fellow psych nerds out there, the famous Harvard professor and behaviorist B.F. Skinner defines this as a variable reward schedule. Skinner observed that mice responded the strongest to random rewards. When mice pressed the lever, sometimes they got a small treat, sometimes they got a large treat and sometimes nothing at all. And like other mice that received the same treat every time they press the lever, the mice that received variable rewards press the lever more often and more compulsively. So, how do you apps leverage this strategy? 

Well, let’s take Instagram as an example. Their notification algorithm is designed to sometimes withhold likes on your phone and deliver them in a larger burst at a later time. So, imagine the time that you posted what you think is an amazing photo and you attach a perfectly clever caption. When you first went back to check, the response was not what you expected. Maybe you didn’t get any likes which led you to feel a little disappointed. Then when you check back again, probably pretty soon after that first check, you see 20 likes. Because of the initial negative outcome, your dopamine centers are primed for a heightened response to the sudden social praise you received in that second look. So, we’re just like Skinner’s mice and we end up compulsively checking for the next big reward, or in this case, the next burst of likes. 

All right, so what does all of this mean to our industry? What is the insurance industry doing about it? We took a look at industry losses, smartphone penetration in the U.S., and the distracted driving activity of telematics users over the last several years and found the percentage of losses attributed to distraction has tripled. So, that’s from 2011 to 2018, the percentage of losses attributed to distraction has tripled and costed our industry about $9 billion. And during this time, keep in mind, so in 2011, we went from 35% smartphone penetration in the U.S. to today where we’re at something like 80% penetration. This is exactly why we at Arity have gone all in on distracted driving for the insurance industry and most large insurers, as we saw even in the survey, are starting to do something about it. There are many awareness campaigns, educational, social, even lobbying campaigns. And, so, we’re getting a start on it. We’ve started really contributing to this conversation, but we really wanted to understand more about how distracted driving is impacting driving behavior and losses. 

We took a look at industry losses, smartphone penetration in the U.S., and the distracted driving activity of telematics users over the last several years and found the percentage of losses attributed to distraction has tripled.

So, we took a look at our own data and we wanted to see what happens in your actual driving when you are distracted, looking at a basic driving behavior—hard braking—around the time that a driver uses their phone. This graph shows users braking behavior starting at the time of distracted driving and we found that immediately after engaging with the phone, your hard braking is 70% higher than your normal driving pattern. And not only that, but it takes about eight seconds to recover back to normal driving. So, for us, we also need to understand the impact to insurance losses. When we look at our own massive mobile data sets, along with the actual losses and traditional factors, we find that distracted driving behavior is predictive above everything else we already know. And in fact, the most distracted drivers have a loss cost more than one and a half times set of the least distracted. So, we can certainly justify pricing incentives to reduce the behavior. 

OK, we know that distraction by smartphones is a problem and we also know that phones aren’t going away. Smartphones, in particular, are not going away. They’ve become the primary means for communication and for completing our everyday tasks. We are addicted to our phones and it is impacting our performance on the road and causing more losses. We also know consumers need help. They believe it is a problem, but they don’t know the extent of their own behavior. So, what are we to do about it? Well, we can use the phone as a tool for good. We’ll start by accurately identifying distracted driving. Now, I’m going to turn it over to Carl to talk about how to use the sensors in the phone to detect distracted driving. 

Carl Schmidt: 

Thank you, Katie. If you take nothing else away from our presentation, just remember that we are no better than the mice. Thank you. So, prevention begins with detection. The good news is with the impressive sensor arrays available on most modern phones, this is absolutely doable today. But there are right ways and wrong ways to go about it. My objective for you is to come away with some of our best practices. So, let’s start with a discussion of general driving risk. How do you identify a risky behavior? When the insurance industry adopted the credit score, it inherited the predictor variables and the definitions of those variables from another industry and it just pointed them to a new target variable—insurance loss versus credit default. That’s not the case with telematics. No one provided us with a list of risky driving behaviors, or instructions for how to identify them from sensor data. 

We had to do it ourselves. For most TSPs, I presume the process of identifying new risks starts with intuition—seems reasonable to posit that a driver who slows suddenly or is distracted pretty frequently is more likely to get into an accident than a driver who doesn’t. The hard part is being able to reasonably and reliably identify that behavior in sensor data and invalidating that it is indeed predictive of losses. It’s hard even if you have all of the data, that is, the telematics data, the traditional risk characteristics and the claims. It’s pretty much impossible if you don’t. We believe new driving risks—I apologize. We believe modeling new driving risks, alongside large volumes of insurance data, is paramount. 

So, I want to take you through some do nots of detecting distracted driving. First, the embedded sensors alone, such as OBD, is not sufficient to detect distracted driving. While these things may be the gold standard for certain behaviors, like hard braking and speeding, et cetera, the mobile phone gives us a valuable tool to understand distracted driving in a way that an OBD cannot. Number two, phone movement only reveals part of the story. You cannot rely entirely on phone movement. Let me give the example of somebody who keeps their phone in a cradle at all times when they drive. This person may indeed interact with their phone a whole lot, but if you are going by only the accelerometer and the gyroscope on the phone, you may fail to capture some of that risky behavior. We need to couple the phone sensors with other information. Number three, discrete events can be misleading. At Arity, we like to think of distracted driving in the form of sessions with a distinct beginning and a distinct end.  

It is the duration of those sessions that we believe drives risk. And, finally, not all distracted driving is equally risky. You need to take context into consideration when assessing distracted driving. If I’m parked responsibly at a red light and I happen to take out my phone to see if my daughter had called, that may not be the same level of risk as flying down the highway, in congestion, and taking a selfie of myself. We need to try to do our best to identify and separate those risks. So, how do we do these things? Well, we generally will begin thinking about the problem with our physical intuition. We can use physics, we can use logic, we can use our expertise, with the hardware and operating systems on the phone, but we need to hone that approach with data. What kind of data would we use for identifying distracted driving? 

There’s a whole bunch of sensors on the modern phone such as the accelerometer, the gyroscope, the barometer, and more. And we can also couple that with information we can get from the operating system such as when somebody has unlocked or locked their phone. And finally, we must combine all of that information with context. So, can we use the GPS to understand speeds at which distracted driving took place? Can we use the GPS lat and long to understand whether or not the person was in a congested area when the distracted driving took place? And, we believe it’s by combining all of these factors that you’ll arrive at the best user experience and the best overall risk prediction. Fidelity of the distracted driving prediction, we believe, will also drive fidelity of the risk prediction. 

So, how are we living into these principles? Well, we have an ongoing program referred to as the naturalistic distracted driving program, and it is kind of what you see here. We’ve equipped a number of drivers with inward-facing dash cams and a telematics application on their phone that captures comprehensive information about what the phone is seeing at all times. The objective of this experiment is twofold. Number one, capture distracted driving data in a naturalistic setting. We could and we have gone out and generated a lot of simulated distracted driving. That’s sufficient to build models, but there is risk in creating bias if you build and validate your models on simulated behavior. Our number two objective, leverage video data as a reliable source of truth. Distracted driving is really tricky. It’s hard to get perfect labels of when distracted driving has taken place and what kind of distracted driving was taking place. 

The camera is one way that we can get that perfect source of truth, but we need a solution that will scale. It is not sufficient to have one, or two, or a half dozen drivers out there collecting distracted driving. So, we have developed a framework that is scalable, begins with the video data collection, but to scale to hundreds, then thousands, then tens of thousands of hours of video data, we have needed to develop an autonomous labeling system that takes the video data and returns a classification of whether or not distracted driving was taking place. And, now, this solution is not perfect. So, we do use manual confirmation to clean up those labels and add additional attributes. And the culmination of this pipeline is a modeling file with comprehensive telematics information and perfect labels for when and how distracted driving was taking place. So, I’d like to spend one slide on the autonomous labeling solution because I, as a data scientist, find it extremely fascinating. 

What you’re going to see on the next slide is a rendering of what’s called the SHAP algorithm. This is an algorithm that helps us visualize what a convolutional neural network is seeing. In other words, it will help us understand the parts of the video that are exciting the convolutional neural network. Go ahead. So, in this video, I don’t have a crazy disco ball in my car. What you’re actually seeing is a rendering of what’s exciting the computer-vision algorithm. The red pixels indicate excitement that convinces the algorithm that distracted driving is taking place. And, now, as you’ll see when I set the phone down shortly, the blue becomes more excited. The blue pixels are those things that are convincing the computer-vision algorithm that distracted driving is not taking place. And what’s really cool here is the algorithm understands the phone, it understands my hand, and it also understands the association between those two things. So, as I pick up my phone, here we go, back to red. Go to the next slide. So, that’s extremely cool, and it allows us to scale up, but it is a means to an end. What we really care about is the ability to predict distracted driving using only the information we would be able to get from the mobile phone. So, I’m going to show a few more videos to help you understand how that might be possible. 

In this first video, you’re going to see me engaging in taking a phone call and you’re going to see how one of the sensors on the phone, the accelerometer, responds to that behavior. And, I’ll let this play with a little bit of silence. You see me aggressively nodding, and, shortly, I’m going to set the phone down, and then set it in the coffee holder, and I’ll let this cycle through one more time for about 15 seconds just so you can see the repeatability of these patterns. Great. So, in the next video, you’re going to see me engaging in a texting activity. So, same thing, you see the first—when I pick up the phone, and, now, I’m in a tapping behavior, and, shortly, I will set the phone down and we see the big spike. And again, I’ll let this play for about 10 more seconds, so you see the repeatability of some of these patterns. 

Great. So, let’s have a look at both of those behaviors side by side. Can we distinguish a call from a text? Well, looking at these time histories, I can pick out a number of different actions that correspond to one behavior versus the other. One thing that jumps out in the case of the phone call, there were two bursts before and two bursts after the sustained behavior. Whereas with texting, because I only brought it up to my hip and not up to my ear, I only really see one distinct burst before and after the behavior. And then within the sustained behavior, there are differences we can look at as well. If you want to jump to the next slide. 

So, if you kind of peer at the data on the left, you might observe some differences, but we can get even more insight into the differences by using what’s called spectral analysis. Let me help you interpret the plot on the right. What this tells us is that the texting and the phone calling behavior were awfully similar in some ways, but the way in which texting is different is that it excites behaviors at a higher frequency. There are more cycles associated with texting than there are associated with taking a phone. And in fact, the big peak that you see out near about five cycles per second and then another one near six cycles per second, that corresponds with something on the order of 60 to 70 words per minute. Pretty good, right? So, finally, you need to have a deployment strategy for your distracted driving algorithms. We believe in a deployment strategy that is modular, that includes a fusion of both OS interaction and movement, includes information to help contextualize the mode of distracted driving and other conditions present at the time, is verified on a naturalistic distracted driving dataset, and is validated against insurance loss data. With these and all the other aforementioned best practices, reliable detection of distracted driving is not a promise of the future. It is the reality of now. So, the critical question remains, can we use this to motivate change? 

Katie DeGraaf: 

Thank you, Carl. Using these sensors to accurately identify what a user believes is distracted driving behavior is certainly the starting point. For us, we then take these complex distracted driving attributes and identify which are actually causing insurance losses. So, by this, I mean we must understand which of these behaviors are not just causing accidents, but we must also take into account the severity, the cost of the accident to accurately identify the least and most riskiest distracted behaviors. This, after all, is the business we’re in, right—identifying how behaviors and characteristics explain the likelihood and cost of a claim. So, in parallel to that, at the end of the day, what really matters is that we’re able to express these back to a driver in a way that is meaningful and relevant in order to create sustained behavior change. 

This is a complex problem to solve because changing driving behavior is difficult. Most drivers have not received objective feedback on their driving behavior since they graduated from drivers ed. It’s also complex because all drivers are not alike. There is no one-size-fits-all approach. Each person is different and the way they want to be communicated with is unique and it changes over time. We like to use millennials as an example. There’s a tendency to bucket entire generations and make sweeping generalizations, such as millennials have beards. Note that I am sitting across the table from Carl, who is in fact a millennial with a beard. But, Carl aside, in the US the millennial population is about the size of the UK. So, building features or experiences to work for a segment this broad is like building one app and saying, “Well, that’ll work for the English.” So, our approach to solving this problem of engaging and motivating drivers to be safer and, in particular, end distracted driving is rooted in a human-centered design and behavioral science in order to really understand human preference at a very granular level. 

So, these teams have partnered together on AB testing, consumer research, and lots of experimentation. And we’ve identified four key approaches that create an engaging experience to motivate safer driving. First approach is just to make it available. A simple access and clear display of driving behavior is all some users will need to become aware, take ownership, and change their behavior. An example of this is my own daughter who has been connected to telematics programs since she started driving. And, from the very beginning, she would daily review the trips reported on her app and come talk to me about the behavior that were reported and what she believed had occurred. She is a far better, safer driver than I am. She’s now away at college and has become a distracted driving advocate. She was given some information and we’ve accomplished the goal. Done. Tactic number two, for getting your drivers to engage with your telematics program is to make it personal, provide the information back to them in terms of displays that are meaningful and relevant to them. 

An example of this is expressing the money saved due to safe driving in terms of new lattes. So, you’re safe driving, saved you the equivalent of two lattes. The third engagement tactic is to make checking the app, or behavior, a habit or part of their routine. Behavioral science tells us that to make a behavior habitual or routine, it’s all about timing. You have to encourage the behavior with a particular frequency and the encouragement has to be fairly frequent, but not too often, or it’ll lose its punch. The encouragement has to come with a bit of a surprise. This is exactly like those mice we were talking about earlier. Engagement tactic number four is to make your telematics program competitive. The behavioral science behind this is pretty straightforward to many of us. It’s important to know we’re doing well, but beyond that, it’s important to know that we’re doing better than other people. But, you have to be careful when you provide a competitive type of experience in app. Competition can hurt as much as it can help and users are sensitive to who they’re competing against or how they’re compared. If you show them the wrong peer group, they’ll get uncomfortable and disengaged, and they’re also sensitive to how their performance is shown to them. 

And I love this quote from one of our app users. I love it. It says, “I now have insight into how many times I pick up my phone in a given trip. On a recent commute home. I didn’t pick up my phone once and was so impressed with myself.” I love it because it’s a micro step and she took one trip—that’s realistic and it’s the first step to sustained behavior change. I just want to thank you all so much for your time and attention on this topic. Distracted driving through phone use is causing too many unnecessary deaths, but the phone can also be used as a tool for good. The sensors can be used to accurately detect behaviors that are relevant to the drivers, and a comprehensive approach to engagement can create sustainable behavior change. I hope you take away from this session that Arity is very serious about this. 

We believe insurers can and should contribute to ending distracted driving. It’s a complex problem to solve. So, we’ve dedicated teams of experts to understand it end to end. And we believe that a reliable and robust distracted driving solution is possible and should be integrated to affect meaningful change. And a naturalistic experimental framework is necessary to develop and validate distracted driving. We also believe in assessing how different forms of distraction explain insurance losses. That is the business we’re in. And, finally, we believe that it won’t be easy, but it’ll be worth it. If we’ll open up the line now for questions. 

Steve Bell: 

Yep, absolutely. So, thank you very much, Katie and Carl, a fascinating presentation. We’ve got some questions coming in. The first question is, do people who have high distracted driving risk also display other driving behavior risks such as hard braking, speeding, et cetera? 

Katie DeGraaf: 

Yeah, good question. 

Carl Schmidt: 

Yeah, we’ve performed a study along those lines where we’ve looked to see how much of distracted driving is explainable by some of those other traditional risk factors like speeding, like braking, like time-of-day mileage. And our finding is that less than 20% of distracted driving is actually explained by those other factors. And I think that tells us that distracted driving tendencies are truly a unique type of risk. 

Steve Bell:

As connected cars through OEMs continue to grow in the market, how will this impact mobile telematics solutions being built today?

Katie DeGraaf: 

Yeah, good question. So, connected cars are all part of this trajectory of future mobility, right? Where consumers are taking different modes of transport, they’re getting to their destination in different ways, they’re driving smarter and more connected cars. And so the OEM is a part of this ecosystem and a part to solving the problem. But, today, and in the foreseeable future, let’s say the next several years, today we’ve got something like 80% of consumers that are attached onto smartphones and, again, causing the distraction in the vehicle, and it’s going to take us several years to get to that penetration in terms of connected cars. So, it’s a piece, it’s a part, but in the end, what do we need to know? We need to know how people are moving and the risks associated with those moves and certainly mobile is a way to do that at scale for the next several years. 

Steve Bell: 

OK, another question. Have you managed the impact or measured the impact of hands-free mobile phone use? 

Carl Schmidt: 

We measure the impact of hands-free mobile phone use. So, we have been primarily focused on those forms of distracted driving that are identified through the sensors and the operating systems on the phone. We’re able to look at different modes of behavior and we’re starting to be able to look at how risk correlates with those different modes of behavior. As far as segmenting the risk of texting versus navigation versus taking a hands-free phone call, that is something that we are investigating actively, but I wouldn’t be able to provide a number. 

Katie DeGraaf: 

Yeah. Thanks Carl. Again, keep in mind, what matters to us is what is the—two things—what is the behavior that’s meaningful and relevant to the consumer? Plus, what are the behaviors that explain insurance losses? So, we dug in on this sensor and OS interaction because that seems to address both of those problems. Consumers, in the testing through the naturalistic distracted driving study that Carl talked through, consumers understand and they accept these driving behaviors sessions that he’s detecting. So, that is relevant to them and that is meaningful to them. And we also know these things do predict losses. 

Steve Bell: 

Next question, how would it help in insurance claim reduction? 

Katie DeGraaf: 

How can that help in insurance claim reduction? I think the primary ways are by getting that engagement back to the consumer, and again, following that comprehensive experience that you give your consumers back that show that taking these things in total really does motivate people to take different behaviors and act differently, thereby reducing their own distracted driving behavior, or other safe driving habits you can form. And then ultimately reducing those actual losses. I think the earlier stat was if 27% of our all losses involve some level of distraction, then anything we can do to give the user back that sort of information—again, we also showed that users, consumers don’t really understand. They don’t really understand how much they, themselves, are distracted and the impact that’s having on the road. So, that’s the piece we’re missing, giving that back to them, so that they can take ownership and change that behavior thereby reducing those accidents. 

Carl Schmidt: 

And I’d also add that we’ve seen solutions on the market that are using in the moment—intervention, such as a chime, or a vibration, or something of that sort to help people know when they’ve been engaging with their phone at unsafe speeds or for unsafe duration. 

Steve Bell: 

OK. Another question is how would you recognize a false alarm like your kid in the backseat using the phone to play a game?

Carl Schmidt: 

We are unable to get through a presentation, like this one, without getting that question.  

Katie DeGraaf: 

Our most popular question.  

Carl Schmidt: 

It definitely comes up every time we talk about this. One thing I’d point out is that when you do make the exchange, that is distracted driving, that probably should count as risky behavior. And, so, we would attempt to identify that moment of exchange. There are sophisticated algorithms that we can use to understand or distinguish between a phone being on the left versus the right side of the vehicle or a phone being in the front versus the rear of the vehicle. And, so, it’s a problem that we are working on constantly to avoid those kind of false positives. 

Steve Bell: 

And another question is why not just use the do not disturb option on the phone and the phone does not ring and the text received, there is no notification? 

Katie DeGraaf: 

Yeah, we—I held up great hope when that feature was released on the iPhone. I know I still have mine enabled, but I also think about distracted driving and driving risk every single day. But, though consumers do turn it off, over the course of this decade that I’ve been working in this space, there’s been a number of solutions that have come out to sort of disable some features or do this—intercept the text or hold it until the driving is finished. And, very consistently, consumers just don’t like it. They’ve had this freedom and this control over their phone for many years now and they just don’t like giving that up. So, if they make the choice, like Ann did, as an example in our quote, if they make the choice of not using their phone while driving, they’re willing to do that, but somebody else taking away that right just seems to—they’re not open to that. And we continually take this back out and test it and ask consumer questions and it’s just—they’re not ready for it. Maybe we’ll get there. 

Steve Bell: 

So, do you sort of keep nudging people is the, I guess, the approach? 

Katie DeGraaf: 

Yeah, I’m a lot of fun at parties, Steve, 

Steve Bell: 

I’ll bet. Along those lines, do you need a completely different set of training data for different markets or, say for countries that drive on the wrong side of the road or the correct side of the road depending upon where you’re from? 

Katie DeGraaf: 

Yeah, so one, we find that pretty consistently across the world, or, at least the regions we’ve been able to test, that bad driving behaviors are bad driving behaviors. The thing that changes from region to region is the distribution of those typically caused by things like accuracy of data collection as well as just infrastructure, road type and structures. And that applies to distracted driving as well. It’s—you see some regions do it more or less, but certainly always risky. 

Steve Bell: 

OK. Another question is, can you explain how the capture and analysis of various kinds of data results in a deeper understanding of distracted driving?

Katie DeGraaf: 

Repeat that for me. Can you repeat that one more time, Steve? Sorry. 

Steve Bell: 

Yeah, sure. Could you explain how the capture and analysis of various kinds of data results in a deeper understanding of distracted driving? 

Carl Schmidt: 

Yeah. I’ll give an anecdote that I think kind of illustrates the point. A lot of—in the early going, say four or five years ago, there was use of the sensors on the phone, such as the gyroscope or the variometer, to understand when distracted driving has taken place. And you could use something like that to kind of look for the movement in somebody, the phone moving from the waist position to the ear position. And, by doing something like that, you might develop an algorithm that’s pretty good a lot of the time, but one common false-positive that you wouldn’t think of off the bat is the motorcyclist who keeps their phone in their pocket and each time they shift their gear, it creates a movement that does resemble a distracted driving behavior. And, so, just the gyroscope is not enough. And, so, we needed to introduce other mechanisms, such as when the phone had been unlocked corresponding with that movement, to help us understand when distracted driving was taking place. And that’s just one example. As you encounter different types of false-positives and false-negatives, you really learn that there’s value in each individual sensor and that the best of all worlds involves the fusion of all of this information. 

Katie DeGraaf: 

Yes, this actually reminds me a lot of the evolution of insurance-pricing sophistication. Just a couple of decades ago, we were pricing based on the data points that we could ask of a consumer, right? Tell us these 14 things. And then we spent two decades learning more and more about the policy, the household, the driver, their history, all about their credit, their shopping behavior, what magazines I subscribe to. We learned all of the things, which then gave us even a better understanding and appreciation of what were the things that could—are correlated with insurance loss. And it’s similar here, right? We’re relying on this technology that’s evolving over time, so we’re getting smarter about what matters, what doesn’t matter. Meanwhile, the technology itself is changing, so we’re collecting new things, adding on new things to the context, just understand what are those actual things that are causing distracted driving and insurance accidents. 

Steve Bell: 

OK. Another question. You mentioned various engagement tactics to improve behavioral change. Have you tested, implemented, or seen impact from them?

Katie DeGraaf: 

Yeah, so each of these we know works. So, any one of these, you’ll get some results, you’ll see some change. I’m sure anybody on the line with a mature telematics program who has frankly just made it available, made the information available, has seen that drivers get better, right? Just, so, you will see some segments that improve on any one of these tactics. So, we do know that they each work for some segment, but again, the point is in order to really address the market and to affect change in a scalable way, in a meaningful way across most of the population, it’s a complete comprehensive approach. We can’t do this simple one solution, one size fits all. 

Steve Bell: 

OK. We’re getting towards the top of the hour. The survey that Becky mentioned at the beginning is on the screen, so I’d encourage people to respond to that and help us make sure that we’re bringing you the best webinars. In the meantime, I’d like to take the opportunity to thank Katie and Carl. It’s been a fascinating hour of insight into how technology and the mobile phone use can be utilized to better improve knowledge of distracted driving, and also, in fact, the enhancement of mechanisms to go there. So, Katie, Carl, thank you very much for your time and your insight.

Katie DeGraaf: 

Thank you, Steve.  

Carl Schmidt: 

Yeah, thank you, Steve. Thanks to the audience.