Bridging the Insurance Talent Gap with AI | Episode 10

Caryn Warner: Hello! Welcome to the Sapiens Insurance 360 podcast. I’m your host, Caryn Warner, and I’m so excited that you’re out there listening. This is where we discuss the latest news, trends, and issues from across the insurance solutions and technology spectrum. Today we have a very special guest, Jeffrey M. Snider. With nearly 25 years of successfully leading teams in cutting-edge technology companies, Jeff is General Manager of P&C at Gradient AI. He leverages his experience managing insurance programs and handling claims as a lawyer, in-house counsel, and risk manager to help Gradient AI grow its solution set to include all major property and casualty insurance lines. On today’s program, we’ll be discussing bridging the insurance talent gap with AI. Jeff, welcome to the show.

Jeffrey M. Snider: Thanks, Caryn.

Caryn Warner: I’m sure that we can all agree that today’s insurance industry is incredibly dynamic. The innovative technology that’s coming out of startups and established firms these days is astounding — in-depth data analytics, advanced API infrastructure, and intuitive and seamless user experience, just to name a few. But a glaring downside is that the industry’s headcount is on a serious decline. According to insurance industry consultancy, the Jacobson Group, the talent shortage is getting worse with the 10-year average unemployment rate in insurance only slightly above the current 2.5%. There are a variety of reasons for the tight insurance labor market, but the more immediate pressing question is what to do about it. One oft-proposed solution has been AI, especially for such key areas as technology, claims, an underwriting, where the retirement of so many baby boomers has been driving a host of vacancies. So Jeff, to start us off, how does AI help insurers who are facing a shrinking workforce and challenge to find experienced replacements?

Jeffrey M. Snider: It’s a great question. I think the answer is to hang on to that institutional knowledge by mining it from the results that those people have achieved, and using it to guide newer underwriters and adjusters and AI allows you to do just that.

Caryn Warner: Okay. So can you talk then about how AI captures critical institutional knowledge? What does that mean and how does it work?

Jeffrey M. Snider: So, if you think about it, we use an analogy that it’s a lot like the oil industry was a hundred and something years ago. They know there’s a tremendous amount of value that they’re sitting on, but it’s just beneath the surface. And the trick is how to access it. And I think if you look at the insurance industry today, it’s the same thing, and it’s being brought to a head as trained industry professionals are retiring and moving out of the industry, and it’s a real challenge for the industry to find people to replace them and to train those people if they can find them. The way we tie that analogy back is all that oil under the ground is all the experience that those adjusters and underwriters have. And the question is, how do you tap that? The way you tap it is you look at everything that they’ve done, all the decisions that they’ve made, and what the outcomes that resulted from those decisions are, and you glean from that what is likely to happen in a similar situation next time. And so by looking at those decisions, what was decided, what was done, and what happened, what the outcome was, and drawing a mathematical correlation between the two, we can make a very accurate prediction about what’s going to happen when we see that same situation again. Actuarial tables allow underwriters to make underwriting decisions, claims managers to make reserve decisions, and ultimately settlement decisions in the aggregate, right? Actuarial tables have been around for hundreds of years, and what they say is that if you have a large enough number of samples, everything deviates to the mean. And so reverts to the mean. And so you know that on average, something like this is going to result in the following. But the problem is, no situation is average, right? The average is made up of things that are better than average, worse than average, riskier than average, less risky than average. But actuarial tables don’t allow you to dig into that. AI and specifically machine learning allows you to evaluate things at the individual level. So at a very high level, the way it works is we train a model the same way an adjuster or an underwriter is trained. We show it lots of inputs. Here’s a policy application, here’s what we knew about it at the time an application was being submitted, and here’s what happened. This is all historical, so we know what happens. And then we train the model to learn and then we test the model by showing it inputs it’s never seen and asking it to predict the outcomes. And then we measure how well it did, and that’s what proves that the model can make those individual decisions.

Caryn Warner: Yeah, it sounds so powerful. So talk to me about how the newer, less experienced and underwriters, claims adjusters, how do they use it, specifically? It captures so much information, how do they access it and how do they apply it?

Jeffrey M. Snider: Great questions. In terms of how it’s accessed, that really depends on the organization. And so the goal, I think, always when you’re trying to do something like this decision support is to integrate it into the decision-making process. And so they can access it in a very simple way, like a dashboard view, but it can also be implemented to drive rules-based decisions. So it may be that a particularly risky application or an application that the AI determines to be particularly risky is automatically directed to a very experienced underwriter or a claim that is determined to be very likely to be high cost is automatically directed to an experienced adjuster. I think the idea is that it should function to supplement or conceivably even replace the experienced adjuster who has moved on or the experienced underwriter who has moved on by providing kind of a virtual tap on the shoulder. In the old days, if you talk to experienced professionals with 20 or 30 years under their belts, they’ll tell you that they received real-life taps on the shoulders from the adjusters or the underwriters who were sitting at the desk next to them. That experienced person would overhear a phone call and say, hey, did you think about doing this? Or when I see that fact pattern, I always do this or that. So for an example, if an expensive surgery has been scheduled, an adjuster may want to have an independent medical exam — IME — to confirm the diagnosis before it’s authorized. An experienced adjuster would know to do that, but a new one might not. And with the power of AI, the new adjuster can be prompted to consider it. Over time, that person will learn to recognize those patterns in the same way they would have before, but now they can sort of do it without a physical person next to them, which in a virtual world is happening less and less.

Caryn Warner: Yeah, that’s fascinating. Does it take a lot of time for underwriters and adjusters, particularly the ones with more experience, to gain an understanding of and trust in AI? And what about other innovative technologies such as predictive analytics? How do insurers help ease what may be, I guess, a pretty steep learning curve?

Jeffrey M. Snider: Yeah, it’s not really as steep as it sounds. And the reason is that we’re not looking to replace industry professionals. We’re looking to augment their capabilities. We say we’re looking to give them power tools. In the 19th century, lumberjacks cut down trees with an ax. That’s all they had, so that’s what they used. It worked, but when somebody came along with a chainsaw and showed them how it worked, they were pretty quick to adopt it. And so it’s really that kind of a process. And what we see is the experienced adjusters or underwriters are the ones who are asking for this the most. The ones at the most innovative companies recognize that this isn’t some sort of magic, it’s math. And it’s math that they’ve been trying to do all along. And again, at the aggregate level, they do very well. But the ability to do that at the level of the individual risk is really new. AI, it doesn’t provide a crystal ball, right? We’re never going to be able to let you see the future. What this is, is essentially card counting on an industrial scale, and it allows an underwriter or an adjuster to see the most mathematically probable outcome and to consider that in his or her decision-making. It can also be used as a sanity check, what we like to call an intelligent audit. Traditional audits are done by first identifying a representative sample of claims or policies, which when you think about it, is kind of a silly way to go about it, because in a representative sample, there’s nothing wrong with almost all of the things that you’re looking at. So you’re looking at a needle in a haystack, looking for a needle in a haystack, by taking a small piece of the haystack. What AI allows you to do is instead say, okay, here’s what your folks think. Here’s what the AI-driven model thinks. Let’s take a look at where they agree and where they disagree, and where they disagree by the most is really where you should spend your time. I use an example often when I’m talking to people to illustrate this, pick the hypothetical 20-year industry veteran. Let’s stick with underwriting. How many policy applications has she seen in her 20-year career? 20,000, let’s say? How many of those does she remember? And I’ll be charitable, and I’ll say 200. I say charitable, because if you ask me to sit down and describe 200 of anything, I might get the first 10 or 20, but I’m not going to remember much after that. So how many of the 20,000 policy applications that she seemed to, she remembered very, very few. And why does she remember those? More often than not, she remembers those because they’re the outliers. They’re the ones where what she expected to happen didn’t happen. Maybe she got burned. And so as a result, human underwriters carry and adjusters carry that bias with them, whereas AI driven models don’t. They look at everything as just a mathematical probability. And so once people understand that it’s not threatening to them, it’s not there to replace them, it’s a power tool. And why wouldn’t they use a power tool, particularly where we can show them that mathematical probability, the thing that they now have to largely guess, we can tell them with that much greater certainty.

Caryn Warner: That is a great explanation, and Jeffrey, you’ve made this really understandable and consumable, so thank you for unpacking such a timely and important topic. To everyone listening, thank you so much for spending the time with us here today. We love hearing from you, so if you have comments or would like to follow us on social media, please reach out to us on our channels. And don’t forget to subscribe to the podcast and thank you everybody for listening! We’ve got more coming, so be sure to tune in next time to Sapiens Insurance 360.