In Pursuit of Autonomous Insurance Jason Pamplin, PhD, Product Manager | May 31, 2018


I’ve noticed a big push these days to get “straight through” processing and I’m guessing you have too. There is a feeling that with today’s technology we will finally be able to ask a few simple questions, query a few data sources, know everything we need to know, quote a compelling and profitable price, and book that business right online. Loss ratio will be acceptable and the expense ratio will be CRAZY!  Right?

As a self-proclaimed insurance technology geek, I can completely understand why powerful, future-oriented, insurance brains all over America believe we’re heading in this direction. It seems to make sense on the surface. IoT (Internet of things) is making data collection easier and cheaper than it has ever been, resulting in a surge in data provider and aggregator services. On top of that, people and businesses are all too willing to give up private information in exchange for convenience and low prices. (Can you say, “Google Maps?”)  New “Big Data” technologies are making the mining and interpretation of all this abundant data possible where it wasn’t before. Wow! If we could get all this together in one place, we could have…

Autonomous Cars

I personally hate to drive. In addition, I am a terrible driver. Those things are probably related. But, as an insurance professional, I am stunned that we even allow cars to exist. Cars have been the leading cause of accidental death for a generation and have only been supplanted recently by “The Opioid Crisis.” Some of the smartest people in the world are working on this, and it seems like a death rate of 30,000 people a year should be easy to beat. Seems like we should have driverless cars everywhere.

But we don’t!  Oh, it’s coming. It’s coming with a vengeance and all the economic brimstone that comes with it. But, breakthrough technology such as this doesn’t come easily. It comes with expense and sacrifice akin to the US space program. The first pedestrian fatality from an autonomous vehicle sheds some light on this. In a recent article by Wired magazine, Steven Shladover, a UC Berkeley research engineer, says that the movement of the pedestrian was typical and steady…

That means the problems could have stemmed from the sensors, the way the sensors were positioned, how the sensors’ data was created or stored, or how Uber’s software responded to that data—or a combination of all of the above.

Wait. So, even when the situation is perfect for the car to respond, it might not? Only time will tell how or why this occurred, but it is an important lesson for those of us playing with “autonomous anything”.

What About Autonomous Insurance?

The failure rate of autonomous insurance programs is high. While I will not name them here, there are a slew of projects that were developed to create a “no touch, direct to consumer” product offering that haven’t had near the take-rate that was anticipated. Why?

Daniel Kahneman, in his book Thinking Fast and Slow, points to a fallacy that our brains go through when making judgements. We overvalue the information we know and undervalue the information we don’t. He calls it, “WYSIATI” or “What You See Is All There Is.” While Kahneman doesn’t speak to it directly, it appears to us that organizations are just as prone to WYSIATI as individuals. At Instec, we have noticed a couple of trends that illustrate this problem: the fallacy that all the data we need is on the internet, and the “getting back on the road” problem.

All the data we need is on the Internet

Give me an FEIN (Federal Employer Identification Number) and we can figure out the rest. It is a bit of an exaggeration but not too far off. Certainly, the growing availability of data is truly amazing. But, self-driving cars shine a bright headlight on the distance we still need to travel. A New York Times article from June 2017, cited some of the current shortcomings of autonomous cars:

  • The ability to respond to spoken commands or hand signals from law enforcement or highway safety employees.
  • Detecting which small objects in the roadway must be avoided.
  • Driving safely despite unclear lane markings.
  • Reliably recognizing traffic lights that are not working.

So, self-driving cars do fine if there isn’t too much data (1 & 2) or too little (3 & 4).

There’s a lesson for insurers here, too. Commercial insurance is a complex business, and we should use all the cheap, ethically-obtained, high-quality data we can get our hands on. But, it seems there is still a place for human intuition.

Once you get off, you can’t get back on

Autonomous cars are mostly being designed for driving on the road. There is some work on off-road options, but they are not designed for sustained driving—just to get over obstacles and things. There is VERY little about how a self-driving car gets back on the road if it ever finds itself somewhere else. I propose two possible options:

  1. Just get out, leave the car wherever it is and pretend it didn’t happen, or
  2. If the car allows it (and maybe if it doesn’t), switch to manual control and never allow the car to drive itself ever again.

Likewise, we’ve seen many insurance companies build portal experiences, either for agents, brokers, or insureds. These online experiences take a lot of effort and most of them are really done well. Most of them have basic risk scoring and underwriting “kick outs.”  But what is the solution when things don’t fit the mold perfectly?

  1. Decline it, or
  2. Send it to our manual process

The myopic pursuit of automation leaves little ability to return the customer to an online experience once they go off-road. The customer expected a seamless online experience but instead is dragged through a laborious multi-day back and forth email exchange.

Concluding Thoughts

There is no question that autonomous cars are going to become mainstream eventually, but recent results have shown that it is going to take far longer and be far messier than first proposed. As of this writing, nearly every major effort has been postponed from the hopelessly optimistic promoted timelines. The two main issues mentioned above are just a few. We haven’t even touched on security concerns, fraud, anticipating other humans’ reactions, and more. There is much work to do. Luckily there are progressive steps we can take that don’t require an all-or-nothing approach.

The same is true of straight-through processing. We can build systems that offer incremental improvements by selectively removing human involvement where it is wasteful or undesirable and streamline human involvement where people offer the highest value. Most importantly, we should enter the field of autonomous insurance with humility. It isn’t all worked out. This is still experimental. Let’s zealously pursue this vision with a steady, measured, and iterative approach.

Jason Pamplin has a vast background in the insurance industry, working as the CIO of Thomco Insurance, IT Director at Markel and the Manager of Engineering Services at Vertafore. He is currently the Product Manager for Instec Underwriting. With his deep experience in IT, and vast knowledge of underwriting functions, Jason brings years of industry experience to the team, and is a key player in guiding IT solutions and deploying them. Jason has a PhD in Computer Science from Georgia State University.