While their goals are lofty, the mobile sensing technology and data analytics CMT has created are both down-to-earth and easy for insurance carriers and agents to employ.
Agency Checklists asked if we get a look underneath the hood to see what CMT does and what our readers should know about CMT and what it offers. Co-founder and MIT Professor, Hari Balakrishnan, was kind enough to answer our questions here in our latest Agency Interview.
Professor Balakrishnan. Thank you for taking the time to speak with us.
I am happy to do it, and please call me Hari.
Could you tell us about CMT and its mission?
CMT is headquartered in Cambridge, Massachusetts. Our mission is to make roads safer by making drivers better. We work with a variety of insurance companies both in the US, and, in particular, Massachusetts, and abroad. We also work with many fleet companies, cellular carriers, auto manufacturers, and government agencies: In short, anyone who is interested in safe driving.
What was your background that brought you to become the chief technical officer of CMT and a professor at MIT?
I am a computer scientist by training and education. I joined MIT as a professor nearly 20 years ago. Since joining MIT, I have focused on an area of research called mobile sensing. The idea is to sense the environment around us using sensors that are on mobile devices like our smartphones.
And in 2004, I, along with my colleague Sam Madden who was also a professor of computer science and the Artificial Intelligence Lab at MIT, got very interested in what mobile computing and sensing and artificial intelligence can do for transportation. And you have to think back to 2004-05. We are talking about before the iPhone and before Android phones.
In 2005, the high-end phones were Nokia phones and Motorola phones. Still, we had this realization that sensors were going to become common on phones, by which I mean positioning sensors like GPS and accelerometers and gyroscopes. Such things can allow us to understand the dynamics of the phone’s movement patterns and the environment around the phone’s location.
We had a lot of academic success with the CarTel Project between 2005 and 2010. It was a pioneering project in transportation because it showed that you could use embedded mobile sensing capability to understand something about traffic, road hazards, and road conditions.
In the CarTel Project, we did something called the “pothole patrol” where we embedded devices with accelerometers, which can sense vibration, in taxis in the Boston area. We were able to map the roadways in the Boston area to identify potholes and bad road conditions automatically. These results were published in the Boston Globe and the Wall Street Journal.
Did CMT grow out of your research on the CarTel Project?
Yes. In 2009, Sam and I started thinking about how to apply this technology to the real world to do some good. Two areas piqued our interest. One was traffic, and the other was safety.
As we were digging into both of them, road safety became prominent in our minds, because at the time, people were talking and doing something about this idea called “telematics.” But their ideas required putting new hardware like onboard diagnostic devices (OBD) or embedded black boxes in cars, which cost quite a lot of money.
We felt smartphones could do a lot of the stuff that people were building dedicated devices to do. We also believed smartphones would drive this consumer curve, since every 12 or 18 months people change their phones and therefore get new technologies, new hardware, and new software. Whereas if you put something inside a car, you do not get to upgrade the technology in the car more than every several years.
So, we took this idea, incorporated the company, and developed and refined our smartphone-based solution, calling it DriveWell. Initially, via our connections inside of MIT, we were able to obtain interest from insurance provides after we created the company, including a very large U.S. insurer that has been our customer to date.
Since starting in 2010, we have grown both in the U.S. and abroad to a point where we are a profitable company with roughly 60 employees in the Boston area. We currently have active programs with over 30 insurers in over 20 countries around the world and millions of users using our services, products, and technology.
What are the advantages of your product for insurance companies?
The biggest advantages are: first, it allows the accurate measurement of people’s driving behavior and, second (and equally important), it helps people become better drivers by providing an unprecedented form of feedback about driving together with behavior-changing gamification and incentives.
The benefits for insurers are in three places.
Number one, a substantial reduction in loss ratios. The dominant causes of driving crashes relate to factors like distracted driving caused by phones and at-risk speeding, as well as more traditional telematics factors like the type of hard-braking patterns that people exhibit. And by accurately measuring these factors and then providing incentives for people not to exhibit those bad behaviors, we can reduce the rate of vehicle crashes.
Second, because the program is smartphone-centric and can run as an app, it provides a much better way to acquire new customers and grow your customer base. And we have done that in a variety of ways, including with driving contests for identifying the safe behaviors of drivers and creating a lot of positive publicity around how to acquire new customers. And the cost of acquisition is very low now because all they have to do is download an app.
And number three, I think many insurers are benefiting from better retention because this product is an enjoyable product to use. The user engagement is very, very good. And what that means is perhaps for the first time, you now have a product that is delivered by an insurance company that people want to use!
We get a lot of positive feedback about the engagement that users are having with their app.
Is this app also a social good, also? Its use is saving lives and reducing injuries to people.
That’s exactly right. That is what drives us internally at CMT; we would like to make our roads safer and do that by making drivers better.
The statistics are very alarming. The way I frame the statistics is that in the next minute, between two and three people will have died and 95 people injured on roads worldwide. So, if we can bring those numbers down, that is very beneficial to the world.
Now, is this app only available through insurance companies or is it also available as a stand-alone app for consumers?
That’s a great question. The predominant mode in which users get access to our solution is via our insurance partners. We think that there is a very natural synergy between drivers, insurers, and CMT. But that is not the only way. We also have an increasing number of partnerships with city governments where we offer it via the city and use the public relations with the city to provide this. We also offer programs via fleet companies and cellular carriers.
A good example of this was the Mayor of Boston, and his office of New Urban Mechanics selected us to build and launch “Boston’s Safest Driver,” which was a driving contest that ran for a few months and had a lot of users get involved with the app. We had a similar program in Seattle, and we are going to be doing such programs in different cities, many of which are part of the Vision Zero program. The Vision Zero program is a government initiative, which many cities are participating in about making roads safer.
How is CMT’s app being used in Massachusetts by insurers?
Among several Massachusetts insurers using our program, Plymouth Rock and Arbella are two good examples. And they do different things with it, by the way.
First, you want to ask, “How is this going to make my insurer’s proposition better?” The premise here is that it makes people safer drivers. So, what factors help us make safer drivers?
Number one, there is ongoing measurement, so people are perhaps more conscious of their driving, which I think is the valuable initial starting point. But what the app does is it provides really good feedback. It tells you, after each drive, what you did right, what you did wrong, and it summarizes this information and gives you a score. It tells you personalized driving tips. And these factors, combined together, are a good self-introspective way to learn.
It is the same principle behind why people use their fitness devices. They look at it a couple of times a day maybe to see what, they should stand more or run more or what have you. And this is like that.
And that is the place where Arbella is using it right now with the Wheel Focused app. They believe that by providing this to their clientele, people will use it to become better drivers. And they have the measurement metrics associated with that.
And I do think it is going to work because it has worked in other similar programs.
Plymouth Rock is doing something a little different. They are combining rewards with driver feedback in the Road Rewards app.
The idea here is that the safer you drive, the more eligible you become for different types of rewards like fuel, coffee, or movie tickets. And our system has an integration already in place with a rewards confirmation company. We can automatically post-fill electronic rewards.
These are all positive nudges to the user to go in the right direction and drive more carefully.
The data is used to provide feedback, but these apps also have a social-gamification aspect to it because you get to compare your driving with other people in your town or other people in the state.
It sounds like you can tailor the DriveWell app to whatever type of program you want and private label it?
We call it white-labeling. And, yes, there is a lot of customization that is possible, and it is very fast. The main thing I would want to communicate is it is a matter of days and weeks to produce a really exciting new app, because of the way we structured the DriveWell program internally there are a lot of bells and whistles that you can configure.
And the process is quick. It’s not like you contract with some company that says, “I’m going to take nine months to produce an app.” We did a lot of the work to design our app in a flexible way so we could do any white-labeling quickly.
We also provide a software development kit that is serviced by our APIs so that third parties could build apps. And some of our customers have a third party or an internal app group building an app using our SDK and telematics service. In the case of Arbella and Plymouth Rock, the apps are white-labeled.
But then if you look at those two apps, they look quite different in the end. So that just shows that there is a lot of flexibility in the white labeling.
Could CMT be a victim of its own success? For example, could an insurance company which did not subscribe to your systems say, “We will write anybody who has a rating from CMT of this or better, and give them a discount.”
I suppose to some degree that could happen. But the big hypothesis that we are validating is that safe drivers are made, not born. Some companies use an OBD device or a mobile app to measure their insureds for some short period of three to six weeks or so. I do not think that idea’s benefits are borne out by data for the majority of users.
The majority of users benefit a lot from continuous engagement and continuous measurement. I think that for the industry as a whole, the promise we bring them that that type of measurement will provide the greatest efficiencies. It is not just about the initial segmentation but a more continuous segmentation.
In fact, that initial segmentation I believe is a little fraught, because many people can discipline themselves to drive very differently in that initial period. But then with the continuous program, people condition themselves, and they get to be better in a lasting way.
I do think you are right, but at the same time, I think the biggest gains are in a longer period of measurement. And that is why engagement is crucial. But of course, we are open. You know, some of our customers do use the shorter model, but many of them use the longer-term model.
I think that in the end, if you want something that, in fact, provides some value to the end user you want them to continue to use the program.
What kind of data points do you capture?
We are capturing position data, which includes the vehicle’s velocity. We are capturing acceleration from the three-axis accelerometer on the phone, and we are capturing gyroscope data.
Our algorithms take this raw data which could be noisy, could have errors in it, and convert it into two things. First, the dynamics of the vehicle itself. We want to understand what the vehicle is doing, not what the phone is doing. And second, it also transforms it into what the user is doing with the phone. Is the driver moving their phone, do they have the phone on? Are they typing on the phone? And so forth. We pick all of this stuff up, and the telematics engine computes a dynamic time series data set.
The data set has a timestamp, the location, the velocity, the three-dimensional acceleration pattern, the pattern of distraction if that’s happening, etc.” From that, we can say to the insurer “Here is what we think the user is doing” and answer the question: “Are they distracted or not?”
This then gets fed into a scoring module. It also gets fed into a machine learning AI algorithm that determines whether the user is driving or whether they are a passenger. It also determines whether they are in a car or some other mode of transport like a bus or a motorcycle or a bicycle or a boat or whatever.
And for the cases where the user is in a car and we believe they are driving, we compute the score and produce the score built off a statistical model that captures how these individual driving factors like distraction and speeding and braking and so on, relate to overall crash risk.
Can your data show insurers if insureds drive in a particular area there will be a statistically significant accident frequency that is independent of how well those insureds drive?
Yeah, that is a good question. We have our scoring models, and then we typically work with actuaries and companies to adjust and adapt our scoring models for different customers. There are some customers who care about, or some actuaries, who care about that factor and it is available, and some that do not.
I can say that for the two specific programs I described above, in Arbella and Plymouth Rock, we do not use that factor. But it is possible to use those factors, like the zip code or location of garaging and things like that.
The purpose of our work in these apps is to take the factors that drivers can control and change it. Like you probably cannot control where you park your car because you work somewhere, you live somewhere, etc.
But there are some of these base levels of risk having to do with the zip code of your home and so forth that already we know how to do that the insurance industry knows how to model. Our apps focus on controllable factors having to do with the way people drive.
There is now a lot of evidence showing that how people drive has a dramatic impact on crash risk.
Do you have any models relating to using this data to minimize premium fraud? For example, some insureds fraudulently claim their vehicles are principally garaged in low premium locations to avoid paying higher premiums?
I can tell you we provide this information [vehicle location] in many cases to our insurance customers who want it. We are in the business of providing accurate information, and I think some insurers may use this information to assess fraud.
Is there any evidence that insureds who will not use your app, might be less desirable drivers from an insurer’s point of view?
I think there are two kinds of people who use the app, people who believe they are good drivers, which turns out to be the vast majority of people. Whether or not it is true, they believe that they are good. The other group is people who don’t think they’re going to benefit from the app.
Certainly, people who are prone to a lot of fraud may or may not want to use it. But, some people prone to fraud may want to use technology like this because they believe they can cheat the technology. And we do have mechanisms in place to assess a lot of this potentially fraudulent activity.
But it is really hard to understand human psychology as to what people believe they can do with technology. Some people believe they can fool it.
You collect a lot of information from your mobile apps what do you do with it and who owns the data from these drivers?
There are two kinds of data. There is telematics data, which is owned by CMT. And there is personal data, or personally identifiable data like a user’s policy number or home address, name, and things like that. And that is typically owned by the insurance company. But all of the anonymized telematics data, the data having to do with location and acceleration and all of that, is usually owned by CMT but is shared with the insurer. It is worth mentioning that the data remains encrypted at all stages and only the insurance provider can access this data.
Where do you see the telematics industry going in the insurance space and what should insurers and insurance agents anticipate as this technology matures?
First off, it is inevitable that technology in the form of data and what we can learn from data will fundamentally, in significant ways, change the face of insurance. I think the other thing that is changing the face of insurance is artificial intelligence (AI).
I think that for insurance companies, in particular, it is going to be critical because AI-based programs like CMT’s DriveWell is being used now by more of the early adopters. We are still in the early stages of this. We are seeing a fast growth rate, but the vast majority of policyholders in the U.S. and abroad are not using telematics.
I think that companies, especially small to medium insurers that are not getting with this technology and with this way of thinking about it, are going to get left behind because they might not be able to come up with accurate ways in which they think about how to price their products.
Not only that, but also how to engage with their policyholders. I think that the consumer experience that you get with an app-centric world is phenomenally better than if you do not have a mobile experience. From that standpoint, it is going to be critical for insurance.
And regarding where the data itself is going to come from, it is just going to be a lot of different sources of data. We think that smartphone-centric and smartphone sensors are ubiquitous and therefore a good source. But data can come from the cars itself. The data could come from other devices. Our DriveWell platform works with a large number of such data sources.
In ten years I think that probably 13-14% of the new cars on the market are going to be autonomous. And we are going to be living in that time frame, in a world where there is going to be a mix of human cars and self-driving vehicles.
Like, when automobiles replaced horses, and the two shared the roads?
Yes. But it is going to be a shared world for a much longer time. Cars last eleven or twelve years for their median lifetime, and many cars on the road run for a lot longer than a decade.
The question after, say, that ten plus year time frame is going to be: “What is the face of insurance in a hybrid autonomous + manual world?” I believe very strongly that the types of technology and analyses we are doing over at CMT are fundamental. Insurance will change from assessing human behavior to assessing the behavior of artificial intelligence and assessing the quality of the artificial intelligence software.
I believe that new autonomous cars are still going to need to be insured. People say that computers and AI will eliminate all accidents. These people are imagining that sensors will be perfect, the AI algorithms will run perfectly, and there will be no software bugs. We are very far from that world! The fact is that we will not be able to eliminate all accidents. We will just be able to reduce them. But there will be different failure modes, and the hybrid world of manual and self-driving may turn out to be more complex than roads today. Insurance will remain important.
But I believe looking further out than ten years, the face of insurance will become much more about data and telematics and AI than it is today. I envision CMT as a company that is focusing on changing human behavior alone but also improving the behavior of AI.
That is the overall version for the ten-year time frame, but it is going to take a while to get there. And during that time, I hope that the majority of vehicles are insured by telematics as we head toward that eventual world where telematics is just going to be the way we do things because every self-driving car will produce a ton of telematics data.
You were talking about the telematics in insurance, are there any statistics regarding the growth of this part of usage-based insurance?
That is a good question. There is an international consulting company, Ptolemus, that conducts informed market research. They do a quarterly report on the usage-based insurance industry worldwide. For 2017, they reported the usage-based insurance market had grown 26% to 17.4 million policies worldwide.
They also identified Cambridge Mobile Telematics as both the number one smartphone-based telematics service provider and also the fastest growing provider in telematics for automobile insurance worldwide.
What about large agencies who work with multiple insurance companies? Are you open to working, creating an app for them or do you want to stick with just insurers?
We are very interested in agencies. In fact, we have had some discussions as to how to do that. And, if you had any advice or lead for us, we would love to follow up on that.
We would be pretty open to working with any agents who are interested in the concept, even if they want just to try it out and check it out. We have a DriveWell app in the store. We will send out a code, and they can try it out. It might also be interesting to see whether we can have a contest between the agents. That is an idea we batted around a little bit. So yeah, anything that your readership gets excited about, we would be happy to explore.
How do companies and agents find out more about CMT and what it can do for their businesses?
They can visit our website to learn more about CMT and the programs we offer. If they are interested in learning how CMT can help their business, they can leave a message on our website’s Contact page, email us at firstname.lastname@example.org or call us at (617) 751-4141. In case they leave a message, someone from our team will get in touch with them within 24 hours.
Thank you so much for taking the time to tell our readers and us about CMT.
You’re most welcome; I enjoyed our conversation and thoughtful questions. Thank you for taking the time to talk with me.