Megan is an experienced insurance professional with over 25 years in the industry. A large part of her career was spent working for insurance carriers in roles ranging from sales, distribution management, product, and digital transformation for underwriters, agents and brokers. She has hands-on experience across several lines including personal, small commercial, mid and large commercial, specialty and reinsurance. In her role at IntellectAI, Megan leads product development, sales, and solutions. Megan is passionate about helping carriers and MGAs improve their underwriting experience resulting in new business growth, improved retention and better underwriting outcomes. Megan has a BA from James Madison University and an MBA from the University of Hartford.
I think this is one of those moments when a technology just pops and grabs everybody’s attention. As big a deal as the iPhone introduction was in 2007, I don’t think I’ve seen anything like this fascination with AI since the first commercial browser was introduced in 1995.
IntellectAI has been working with AI for a long time. How would you frame what we’re experiencing?
AI is not a new hammer, and we’re not just walking around looking for nails, but AI can be another tool in the toolbox. Carriers need to start thinking about where within their workflow they're going to take advantage of it.
I don't see a scenario any time soon where people will look to automate underwriters. Carriers believe that “no one can do underwriting like my underwriters can,” and I agree. But there are parts of the process that we can really start taking a look at.
It’s interesting to go back and look at the end-to-end underwriting workflow and to reevaluate all the places where there’s friction. To this point, we’ve accepted a lot of friction, as if to say, "That's just how insurance works." We didn't have a tool to resolve that friction, but now AI brings something else to the table. We get to revisit the process.
That's why I think the industry is really at the cusp of being transformed.
For instance, we use our embedded AI to do data extraction from a submission. We take everything the agent or broker said about the risk, then we take the carrier's guidelines about what they’d like to write. You put the two together, and you can start looking at any risk. Here are the positive attributes, and here are the negatives ones. Here are the ones in the middle. When you bring the underwriter in, everything is ready for them. They're not having to go look for information.
Underwriting rules can be great, especially in small business, where a lot of the decisions are binary. You don't need AI if there was a claim in the last three years and the guidelines say a submission has to be claim-free. You’re not going to write it.
But when you get to those meatier accounts, there's almost always going to be positive and negative, and the AI can bring to an underwriter’s attention those things of which they should be aware.
The underwriter can then figure out, "Is there a way for us to make money on this risk?" How do we craft terms and conditions and pricing? The underwriter gets to spend the majority of their day on the parts of the process that require a real underwriting skill set and not on mundane tasks like gathering information.
AI can also provide guardrails that help newer underwriters make sure they’re looking at all the relevant aspects of a risk. AI can help seasoned underwriters, too. They know the rules, so they aren’t looking for changes, but you can alter a guideline online and have it hit everybody simultaneously.
As you take this newly possible, end-to-end look at underwriting, are you seeing other friction points that you can address?
One point of friction for brokers and wholesalers is to ensure that the carrier did what they proposed. What was bound? What was issued?
We can compare what was issued versus what the carrier said they were going to issue. Is there anything extra in there? Is there anything missing?
There are organizations that will do this work manually, but our AI can do it faster, better, and cheaper.
That's interesting. Both my brothers were professional poker players. And they would keep track of that sort of thing. How many hands did they play [as opposed to folding immediately]? How many times did they get into the final round of betting on a hand? How many times did they win? They would go back afterward and review their play. Was I too aggressive today? Was I not aggressive enough?
It sounds like the work of an underwriter becomes much more interesting when you take a lot of the mundane work off their plates.
You can start upskilling your underwriting assistant staff, to put them on the journey to underwriting. You really create a career path by getting rid of some of the work that is more clerical. We think the change will make jobs more satisfying and ultimately attract more people to insurance, especially young people.
ChatGPT and the other large language models are letting you communicate more easily with AI than you could before. How does generative AI change things?
As an insurance carrier, you'd love to have documentation on every risk that comes through the door because you are going to see that risk again. But when an underwriter prioritizes their work, documenting the accounts they did not write is a less than desirable task. We can start using AI to do that documentation and provide a summary. When the risk comes back the following year and a different underwriter picks it up, they can get a rundown.
What is the risk? Why did we not write it? The documentation helps determine what to do with it this year.
A lot of information gets lost today, and that's where embedded AI can help.
What are some other opportunities?
I was just on the phone with a prospect. We're going to do loss run extraction for them using embedded AI. Where in the process do they want those loss runs extracted? With today’s manual processes, someone only pulls that information if a decision has been made that at least they want to quote the risk. But would there be value in doing it at the beginning of the process, extracting loss information on risks that you would have weeded out? What could your actuaries do with that data? Could their loss models be different? Could their predictive modeling be different if we were able to provide them loss data on every submission that comes to the door?
When all the data is being manually keyed in, you're not going to get all that information; not for a quick decline. But maybe it's a year later, and you start thinking about getting into a particular class of business, or a particular line of business, and you wonder, how many submissions would you get? What would the losses be? How would you need to price it? Now you have historical data to use for evaluation.
I think this part of the extracting is what you would go to a vendor for, but the carriers then can create their own special sauce. What do I think about all that lost data you extracted? What do I think about my appetite for risk? What do I think about my underwriting guidelines? What triage models should I apply to determine what risks I want to absorb or avoid?
That's where you can say to the technology folks at the carrier that there is super-high-value work to be done. These decisions are literally what make the carrier tick.
What is holding up adoption of AI?
AI tools are going to pop up everywhere. I was just reading an article discussing an assortment of random AI tools, and one picture showed a street sign in New York City in front of a parking spot. It was really a number of signs about rules around when it's a bus stop versus when you can park here and when it's a towing zone and when there is alternate day parking or street sweeping, etc. Someone built an app where you can screenshot the signs, and it can tell you when you can park there.
I got a parking ticket this summer in New York. My husband and I just stood and looked at all the signs, and we thought, "Can we park here?" We decided we could. And we were wrong. That app would have been great to have.
But how would I have even known that app existed?
When I think about all these tools that are coming up in insurance, the other part that's super critical is, how do you tie them into your underwriting workbench, so they’re part of your underwriters’ flow? Non-underwriters love to invent tools for underwriters that we think will make their life easier. Especially when tools are newer, they're almost always piloted outside the underwriting workflow. But the result is that you give underwriters a tool that stops their workflow. You're asking them to do something different, sometimes log into something different, something that they're skeptical about, and then pull the result into their workflow—and hope everyone's doing it the same way. You have to make sure tools are part of the flow, not an afterthought.
When the time comes to take a look at the risk, the app is already there, and it's ready for you and it's part of your process.
I love that app that answers the question, "Can we park here?" I lived in New York for 14 years, off and on, and parking is really confusing. If you're an underwriter and something pops up and says, don't write this risk, or think about this, and it’s part of the process, not a separate tool, well, I can see that being really valuable.
We're not at the point where AI should be making decisions. But what if something is just emerging as a risk, and the AI notes for the underwriter, "We should keep an eye on this. We've had some claims in this area"? It's not a process. It's not a guideline. The AI isn’t telling you what to do. But it's asking you to notice something you might want to consider in your decision making.
This has been a fascinating conversation. Thank you for your time, Megan.