Part II: Why insurers need to monitor, measure, and manage driving behavior
This post features Part II of a recent conversation Arity President Gary Hallgren had with Risk Information Editor Brian Sullivan at the recent Auto Insurance Report National Conference that examined the state of distracted driving today and discusses how insurers can derive insights from mobility data to make roads safer for everyone. Below is a condensed version of the conversation, edited for clarity.
Brian Sullivan: Makes sense. And now let’s circle back to the data sources. One of the nice things about the dongle and the OEM data set is that, while it may not be perfect, it’s much closer to perfect than the phone – the minute the car goes on, you can capture the data and you know exactly what is going, much more so than with the phone, which is more of an approximation. How do you see the shift from the hardwire data to phone-centric data?
Gary Hallgren: Dongle and OEM data do not tell you exactly what is going on. Dongles and OEM data (at least at this point) can’t provide a clear signal of distracted driving the way mobile data can, and distracted driving is having a material financial impact on the insurance industry and material human impact on society. For example, when a driver leaves her primary vehicle and continues with her multi-modal life, she doesn’t take her OEM telematics or dongle with her but she takes her phone everywhere; that’s critically important to understanding risk holistically, and to understanding customers and anticipating their needs.
That all said, the biggest problem we’ve seen in the industry relative to all of these data sources is the relatively common practice of using a driving score model built, trained and tested on device/OEM data to score data from a mobile phone. That is a bad idea, and those pursuing it may find out the hard way. As we’ve discussed, data from these sources are fundamentally different; you can’t just swap one out for the other and rely on the outputs of the model. We know – we have an enormous amount of both dongle and mobile data, with all of the associated claims data. Those who score mobile data on a driving score model built off of device data are likely to price less accurately than they would if they didn’t use any driving score model at all.
BS: So, you can’t apply mobile data to the hardware data model?
GH: You can try but it will not be as accurate. Will you learn something? Sure, but as we’ve discussed, the data coming from these different sources are fundamentally different. Mobile phones provide access to GPS, accelerometer, gyroscope, barometer, etc. from a device that might be in the driver’s hand (we hope not), her pocket, the cup holder, the passenger seat, or even her child’s hands.
You cannot process and interpret mobile data the same way you do dongle/OEM data; they may be trying to tell you the same story but they’re doing it in a different language. Scoring mobile data on a model built from device/OEM data implies an array of assumptions that just aren’t accurate.
BS: So, when you look at distracted driving and you’re working with your modelers, you’re looking at distracted driving and then at driving behaviors like speeding, braking, corners, etc.? Weigh for me the impact potential relative to loss.
GH: Relative to the data sources, it’s critical to understand that they’re all important, they’re not correlated and they’re both going to get better. There was a point in time when the device model was better than mobile, and one could argue that the mobile with distracted driving is probably better than a device. But now we’re also learning things in our mobile data that can bring it back to the device, and insights from device data we’re can bring back to mobile. So, to me, it’s always an evolution.
BS: So, there’s likely to be hybrid moving forward. If GM (data directly from the car) is going to give you ground truth, and the phone is going to give you a distraction, you can put the two of them together?
GH: Right, and this is where Arity sees the future. We’re not an insurance pricing house. We are building a software company that’s focused on transportation for the people who want to identify and quantify risk. We are focused on shared mobility companies, insurance companies, auto companies.
If I’m running a car-share program, for example, I want the device data because I want to know what’s happening to my car because it’s being used in California today and I’m not there with a mobile phone inside of it. While it’s likely that the driver has an app in it, companies want to be assured they can get into the cars and know it’s being driven. As a result, I think it’s only going to be a hybrid.
BS: I think it was 14 years ago I had the first telematics person up here (on stage at AIR) and those that have been here a long time know that I predicted ‘this is going to happen tomorrow.’ I completely got the time table wrong. There’s going to be so much more to do.
GH: I wouldn’t say it that way. The reason credit data moved fast was due to the fact that the raw materials were available. Everyone got on board as long as they could delineate losses – if they could build this insurance model, they could begin to price with it. Subsequently, the industry’s direction has been more focused on providing a discount in order to use a device, which is essentially just pricing somebody that you already have.
So, the change is really oriented around the availability of raw materials at new business. We’ve created a marketplace that gets to the root of where the data is – which is what it’s all about. For example, today we have a relationship with Life360, a family driving app that has 15 million users. And if you’re an insurance carrier and you have 5 percent market share and you are not in telematics today, I would argue that you are, because Arity has 500,000 of your drivers today that we’re collecting data on, measuring their distraction to create an insurance score that is filed in 38 states. We could match that data for you today and you could start to incorporate into pricing for those users. Again, it’s this idea of going to where the data is.
Now, if you want to advertise to these same users today, it’s a similar scenario. While you may not be in telematics today, you may still want to advertise to the good drivers, your ideal drivers. Today we have about 10 insurers advertising on the marketplace. And while we’ll continue to add more publishers to this marketplace, you can start getting exposure to your target audience within these 15 million users. This is where we believe the potential for hybrid data is going to skyrocket – as with OEM data. We, Arity, are going to where the data is, which is where the mobile app publishers are and are starting to score using these inputs.
BS: Progressive would never say “We’re sorry we spent all that time, effort and energy to get Snapshot up and running,” but they are aware that the people who got out in front are winning. I suspect the barrier to entry for telematics would be complex because it will be less company-centric and more marketplace-centric.
GH: It’s not going to be as hard as you might think. We know that insurers prefer to build their own models, but unlike Progressive, most don’t have the raw materials to build one that accounts for distracted driving.
If you’re serious about pricing accurately if you’re serious about accounting for distracted driving, then the data you work with MUST account for distracted drivers, and, more importantly, must be tied to claims. A model will only be a good predictor of insurance losses if it incorporates loss data.
And while initially, insurers may be thinking how hard and long it will be to build the programs that attract customers to get the data they need to build these models, their customers – current and future – are already sharing driving data with others, such as mobile apps like Life360. What you need already exists – insurers can partner with us and leverage the connections and associated loss data we’ve accumulated to stay competitive.