Who needs IoT? Smartphones are the new tracking devices

Arity, a startup spun out of Allstate, uses smartphone apps to collect data that once required dedicated telematics devices

Who needs IoT? Smartphones are the new tracking devices
Credit: Thinkstock

We’ve been told that in the not-too-distant future, IoT devices will quantify our entire world. Sensors will deliver data on every machine, so it can receive maintenance before it fails. Our vital signs will be monitored constantly for anomalies that suggest looming health problems. Already, insurance companies track customer driving habits through telematics to reduce (or increase) rates.

But in an increasing number of cases, dedicated IoT devices aren’t necessary to gather this data when the good old smartphone will suffice. Recently I interviewed Gary Hallgren, president of the Allstate spinoff Arity, who explained why dedicated telematics devices that plug into your vehicle’s onboard diagnostics port may no longer be necessary to collect the data he needs:

Probably three years ago we realized the things we could do on mobile phones. Instead of taking a device and plugging it into the car, maybe we could generate similar kinds of insights -- and it was a much more cost-effective than an actual device. I have to ship it to you; you have to plug it in. Now you could just get the app, load it on your phone, and generate a similar kind of insight.

Telematics devices deliver precise information. Could an app on a phone come close to that? Although the company collects data from a mix of telematics devices and smartphones, Hallgren offers that “in some ways, you can actually get better data” from the latter.

It all depends on what you intend to measure. “The original notion of using a smartphone to replicate what a telematics device can do was flawed," he says. “I have a different set of sensors on the phone than what I have inside of a device that I plug into my car. How do I use the best of what I get from the phone and create a different model?”

Smartphones, notes Hallgren, have accelerometers, GPS capability, barometers, and other sensors. Bad behavior such as jamming on the brakes is easy to detect. “I know if you’re opening and closing your phone during the drive,” he says. “I know that you’ve actually picked your phone up and are looking at it versus having it on the backseat.” One of the toughest problems is to determine when one activity stops and another starts -- or to identify when someone is actually in a car as opposed to riding public transportation. Arity’s machine learning experts are working toward making this sort of pattern recognition highly accurate.

Although wholly owned by Allstate, Arity has opened its platform to external business partners and even competing insurance companies. There are plenty of opportunities to sell Arity’s data, Hallgren notes, including to car companies:

Even though OEMs might know to some degree how their vehicles are performing, I’m wagering that I have a much better idea of how often the vehicles are driven, how many go in for service. I know you came in every 25,000 miles, but I know that actually you take typically five 10-mile trips a day where you drive twice 50 miles a day. I understand how your car behaves. I know what kinds of roads you’re traveling on based upon the average speed of where you’re going. I definitely have a very unique set of insights and you combine that together with what we know as a corporation of who is buying which vehicles, how brands are changing over time, that people are trending more toward this type of vehicle versus another type of vehicle. Now I’m starting to blend all that car-buying behavior together with all of the car-usage behavior and I’m sitting on a really interesting opportunity.

Customers outside the car insurance industry seem necessary in the long run, because, ultimately, the self-driving-car era will change insurance models drastically. Fewer accidents and shifting liability from individual drivers to companies will drive down revenue.

But if one law of computing persists, it’s that huge quantities of detailed behavioristic data can always be monetized. As machine learning grows exponentially better at recognizing patterns in that data, its value will only increase.