I was introduced to Alan by a McKinsey partner in 2007, my first year in the insurance business. Alan had just retired as Group President of Progressive Direct, where he grew the earned premium from under $1.3 billion in 2000 to over $4.3 billion in 2006. I was in awe and even more amazed when the first question he asked me was about baseball.
The following is a discussion about Alan’s insurance journey over the last 40 years – an inspiring reflection by an insurance visionary.
I only asked Alan three questions:
- What did you do at Progressive?
- What are your thoughts on auto insurance and the insurance industry?
- How do you see property and casualty insurance evolving over the next decade?
We split the interview into three parts – Alan’s insightful response to each of these three questions.
——————Part III – The next decade 2020–2030 ————————-
Here is the third and final part of this three-part interview, discussing how Alan sees property and casualty insurance evolving over the next decade.
David Schapiro: Alan, now that we’ve discussed your time at Progressive and some of your insurance industry insights, could you please share your thoughts about the future of property and casualty and auto insurance over the next decade? What will be the key factors influencing the industry through 2030?
Alan Bauer: Going forward, I see three big things going on. One is artificial intelligence/ machine learning, which can and will change the way insurance rates are set.
Another is that consumers are starting to become aware of the data world that’s out there. There are literally hundreds of companies that provide data to insurers, advertisers and others. Things that you never thought of are being tracked and collected and might be provided to others. The last factor is the ubiquity of sensors that are all around us. You can’t go shopping or walk down most streets without being continuously photographed.
These things have effects and will have greater effects over time, whether we like it or not, on insurance.
Auto insurers historically have set rates using algebra, where things are added or multiplied. We ask you your age, and that gets a factor. We ask you your auto type, and that gets a factor. We ask you your driving record, and that gets a factor. All this gets thrown into a generalized linear model (GLM) to output a rate. GLM requires large-scale analyses; for instance, to assess how much more to charge for a Mercedes than for a Ford. You do that independently of whether the driver is a female or male. As a result, you have a relatively easily calculable method of determining a rate. In most states, rates are required to be filed with the regulators, so there is a benefit in simplicity.
Artificial intelligence (AI) may provide a better pricing algorithm, but since we can’t explain how it works, how do we file it with the regulators? We will end up having two different ways of pricing. We have GLM, which is formally going to be the price, and then we have AI, which might be a better predictor but does not lend itself to GLM presentation or explanation. GLM may say that a specific Ford driver is a risk that should be priced at $1,000, and AI may say he should be $500.
Let’s say you’ve tested both and found out that GLM isn’t as good a predictor as AI, but you can’t explain (even to yourself) how the AI calculated the price. You know a particular applicant is really a $500 risk despite the $1,000 price, so you want to get him as a customer. Even though my stated price is $1,000, I want to go after him to insure his car, home or business. I’ll do all sorts of things to try to get him to come to me: call him, send him frequent letters, etc.
What do you do if it’s the other way around? What if GLM says a specific Mercedes driver is a risk that should be priced at $1,000, and AI says $2,500? Would you try to not acquire that customer, maybe putting her on hold when she calls? Maybe ask for specific written documentation – tasks that she will (correctly) view as a hassle and therefore go elsewhere?
AI will play a huge role in acquisition too. Hypothetically, take a case where our AI system says that young men with speeding tickets who drive Fords are actually very good risks. From this, one could use the data providers to find young men with speeding tickets who drive Fords and target the advertising specifically towards those people. I think we’re going to see more and more of that.
I think that AI can combine many different things that you and I would never consider. AI could hypothetically find that red cars with drivers who want to buy the policy on a Thursday afternoon are bad risks. It might not make intuitive sense, but that data and result could be very clear, and that’s the kind of thing that the smart insurer in the future is going to be paying attention to.
Similarly, the sensor data that is out there can predict loss in ways that you can’t with GLM but can with AI. Sensor data is rich AI food and will yield far better predictions of claims, sales, retention and problems.
I think that the future is going to show us who’s good at AI and who’s good at using AI, which are two different things. If you found that the speeding Ford driver is a better risk and you went after him, it could be hard to argue against the fact that that’s bad for society. If you found that the Mercedes driver is a bad risk, and you put her on hold, that may run into some regulatory issues. These are things that some smart companies might start to work on.
In the last 20 or so years, we have had the addition of usage-based insurance (using sensors that track when, where and how the vehicle is operated), which has been, I would say, a good but not fabulous addition to the GLM system. That will grow as the mechanics work out. But as it grows, you will need to discover and avoid, for example, customers with software that can fool their smartphone into not telling the insurer when they’re making phone calls or texting while driving.
Alan Bauer – Bio
Alan Bauer has played a substantial role in changing how automobile insurance is sold and underwritten. His work led to the industry’s first website, its first use of credit scoring in rating automobile insurance and its first internet-only sale. His Progressive tenure included many positions, both line and staff, and assignments in both the agency and direct sides of the company.
While President of Progressive Direct, the company’s direct business earned premium grew from under $1.3 billion in 2000 to over $4.3 billion in 2006, while consistently beating target margins. By the time Bauer left Progressive, the company had become the nation’s third largest auto insurer.
He was awarded some of the first patents for both internet insurance and usage-based insurance. In 2011, he was named as one of the Top Ten Innovators of the Decade by Insurance & Technology magazine.
Since leaving Progressive, Bauer has served on the Board of Trustees at Carleton College (2006–2014; 2017–present) and on the Wikimedia (parent of Wikipedia) audit committee (2007–2008). He has also been active as an investor and consultant, including as the External Senior Adviser on auto insurance to McKinsey & Company.