In this episode of "Exchanges," a podcast by Hitachi Solutions, host Brad Koontz talks with Jonathan Yundt, Industry Director, and Stuart Morris, Director of R&D, about the use of generative AI in the insurance industry. In the insurance sector, generative AI is used in various areas including marketing, sales, underwriting, operations, and claims.
By conversing with data, insurance companies can analyze success rates based on different variables, speeding up processes such as claim processing and policy renewal.
global.hitachi-solutions.com
In this episode of "Exchanges," a podcast by Hitachi Solutions, host Brad Koontz talks with Jonathan Yundt, Industry Director, and Stuart Morris, Director of R&D, about the use of generative AI in the insurance industry. In the insurance sector, generative AI is used in various areas including marketing, sales, underwriting, operations, and claims.
By conversing with data, insurance companies can analyze success rates based on different variables, speeding up processes such as claim processing and policy renewal.
global.hitachi-solutions.com
Brad Koontz 0:11
Hello, welcome to Exchanges, a podcast by Hitachi Solutions.
My name is Brad Koontz.
Welcome to our podcast.
Today we're talking about generative AI use cases for insurance companies.
So generative AI is in the news a lot these days. For those of you who have who have not been catching up on it, it basically is a type of artificial intelligence that creates new content such as text and images by learning patterns from existing data.
So it's been making a lot of news now, not just because of its the fact that it's new tech and has new use cases, but because of the results that it's driving.
According to McKinsey, companies are seeing a 40% increase in productivity through AI driven automation, and it seems like the companies leaning into this are not only seeing major increases in market cap, I just have to look at Microsoft, NVIDIA, Meta and the last couple of cycles.
But individually, we are seeing real productivity gains and that's what we're here to talk about today.
I have two fantastic guests from Hitachi Solutions to talk about how insurance companies can make the most of generative AI.
Jonathan, please introduce yourself.
Jonathan Yundt 1:28
Yeah.
Thank you, Brad.
Jonathan Yundt, Industry director, focused on the insurance industry.
Happy to be here.
Happy to get the gang back together again.
Brad Koontz 1:37
Absolutely, absolutely.
And Stuart, thanks for joining us today.
Stuart Morris 1:42
Thank you, Brad.
Stuart Morris.
I'm the director of R&D, so for the last 12 months that's been all generative AI all the time.
Brad Koontz 1:50
All the time.
I know you're you're.
You're living it every day.
So insurance, insurance leaders today are getting flooded with information around AI around generative AI.
So, Jonathan, what are some of the considerations that insurance companies specifically should have when starting the generative AI journey?
Jonathan Yundt 2:12
Yeah, and we're getting flooded.
Our inboxes are getting flooded with requests for those very same topics, Brad, and the way that I think of Gen AI is is this.
It's twofold.
It's either summarizing something or generating something, and so a lot of our clients were advising them, you know, in our regulated industry, how do I contain the risk?
I wanna use it. I want my employees to use it. A lot of our colleagues and competitors are using it. How do I contain that risk?
We've seen enough horror stories in the news to know that, Gen AI can be used for good and also can chrow some red flags out there.
So how do I contain the risk?
Number two, where do I start?
You know the best way to eat an elephant is 1 bite at a time.
So small bites start slowly and really think about how can you incorporate this into your overall strategy.
It is, as you said, Brad, it's a productivity gain, but it also there's a lot of people who are, if you're not using it, you're gonna be part of, you're gonna be left on the desk.
So we we definitely wanna get our clients started with using it.
And then lastly, how do I control what's going in and what's coming out?
So how do I control the inputs to that we feed the model?
You know you're having employees input data or input prompts.
How do I contain what's going in and then on the on the other side of that equation, how do I control what's coming out? How do I limit my risk and exposure for what is put out into the ether and the public domain and through Microsoft technology we have secure ways to endeavor that that request.
Brad Koontz 4:15
That's a great point, Jonathan.
I think that you touch on a very important topic there in general about generative AI and that's using it, you know kind of the good versus evil use cases that we've got there and making sure that we're using it ethically but also securely.
So talking about the context of containing risk, Stuart, can you talk about the security implications of generative AI?
Stuart Morris 4:39
Uh, sure.
Yeah, we definitely heard a lot out there in the wild about people that haven't been using AI in a secure and and responsible way.
Microsoft has launched their own versions of the Azure Open AI models.
You can get the same exact functionality that's available from open AI directly, but unlike the free tier open AI endpoints when you're talking to the Azure open AI endpoints, there's no training on your data.
There's automatic harmful content moderation built in, so you get a much better sense that you're your business users aren't going to be exposed to things that you wouldn't want them to be exposed to, and you know you're not going to be leaking out sensitive corporate data.
Brad Koontz 5:19
So I wanted to just double click on that a little bit.
Stewart, when you say that there's no training on your data, can you give us a little more idea about there?
Because I think one of the benefits that people see with generative AI is the fact that that it's learning from the data that it's the datasets that it's working with.
Stuart Morris 5:38
It's true.
You can actually go through the process of fine tuning specialized models and the ability to do this sort of traditional artificial intelligence machine learning models that are built for bespoke use cases.
You know, it's been around for years.
One of the interesting things about the quality of these large language models, such as the GPT 4 model, is that they were trained on so much source data that they can actually accomplish the tasks that you want them to do without any specialized training.
The only thing you need to do is supply them a little bit of customized prompting through a process that people are calling prompt engineering.
Now you know with something like a couple of paragraphs of description, you can get them to complete almost any task that you want them to do.
Brad Koontz 6:20
Yeah. That that's great.
Let's talk to Jonathan's other point about input and output.
You just touched on that a little bit related to prompts. Can you just expand on why inputs and outputs are so important?
Stuart Morris 6:33
Sure.
Yeah, we hear a lot of people talking about prompt engineering and this is definitely something where, you know, if you're working with the the kind of chat GPT window where you're having a conversation with the bot, you know that just typing in some stuff is kind of how you get the response out that you want for. For our corporate customers this this isn't enough.
A chat interface isn't really what they're after.
What they're looking for is a way to build a pipeline, sometimes custom pipelines that can get data out of other systems that they use today and get that fed into the bot as well.
So you know, combining some prompt engineering that that goes on behind the scenes along with the data pipeline that can make sure that we get the relevant correct data that it's sanitized properly and formatted the way that the bot can understand it and and generate the correct output from that.
Jonathan Yundt 7:19
You know, there's the time old adage about data; junk data in, junk data out that really speaks to data quality as well.
Obviously it's only as good as the source data that's being fed and there Stewart.
So I I think that that is another big consideration that I that I probably shouldn't put out front is where you getting your data.
That's very much a consideration as part of that input is where where's your data?
What is the status or state is your data in. Is it ready to be to be consumed by an LON?
Is it ready for those very, you know, unstructured or you know queries and prompts that your users will be hitting it with?
I think that's a big consideration as well.
Stuart Morris 8:06
It's true.
Yeah, if you've if you've gone through the exercise of synthesizing data sources that you have today, building out your data warehouse, making sure that things are cleaned, accessible organizable and you know, if you have ways of searching it today, you can bring those to generative AI pipelines if your data is a mess today, it's still a place where you're going to struggle.
They're not going to solve that problem for you.
Jonathan Yundt 8:27
A bit challenging, yeah.
Brad Koontz 8:30
So Jonathan, that's that's that's a really good point as well.
Can you give us some really specific industry examples where we've seen actual productivity gains and this is a place that I get pretty excited about with Gen AI because I think about a lot of different industry use cases across industry, and I really just on the surface, and if you've, you know everybody listening to this podcast is a customer of insurance of somehow, so you understand claims processing and you understand the pain points related to that and you understand what it's like to get a policy at the very beginning. So we're in the, I would say the value chain of insurance firms. Where are they getting actual real productivity gains, use cases.
Jonathan Yundt 9:17
It's everywhere, Brad. It's from, from soup to nuts.
It's, you know, in marketing it's in, you know, the sales and underwriting process.
It is an operations and claims and it it's all pervasive. To give you an example, we've worked with a number of insurance clients that that are using generative AI,
these prompts that Stuart had mentioned, to talk to their data and it's talking to claims data. It's talking to operations underwriting. Help me analyze my you know, success.
My win loss ratio based on different combinations of limits and based on you know premium amounts help me.
It's really helping them hone in on how can I be more successful in what I'm delivering in terms of, you know, whether it's an RFP or whether it's, you know, in underwriting process there, renewals are, you know, big focus of the industry, obviously hitting on captured business, help us help us get to that end result.
Help us find that winning combination a little more quickly. They're actually conversing with their data. The end user is none the wiser.
You know, there it is, as Stewart said, it's synthesizing these massive amounts of data and then presenting a, you know, bolted or streamlined approach, and that that is that's another option too, is that you can have these then output into some sort of proposal document, whether it's a PowerPoint, whether it's a word, you know, some sort of template you can actually have these, you know, dump into a synthesized, formatted way that helps me as an underwriter as the salesperson presenting this information to our clients.
Brad Koontz 11:04
Yeah, that's a that's a another great point.
I think one of the things that you, you know alluded to there was these fantastic use cases on one hand.
Yeah, I think the end customer is none the wiser that this is happening and in some scenarios, but there's got to be other scenarios where they're appreciating the speed increases that have to happen through these processes because they're going to get to it quicker and they're going to get through something like a claim in a much quicker pace because there's the steps become automated very simply.
Jonathan Yundt 11:24
And a gap so.
And that.
Yeah.
And I mean that people, clients, whether you're in personal or commercial insurance clients, remember that you know the speed to value and efficiency.
And so the communication obviously to us, we see this is kind of closing that gap and shortening some of that and then you wrap automation around it Brad and it's really we see you know the transformation in the way that the insurance clients we work with are engaging with their clients. The end user is the policy holders.
Brad Koontz 12:09
Umm.
Jonathan Yundt 12:10
It's more communicative.
It's more collaborative and it's also closing that gap of how long does it take for us to respond?
How long does it take for us to engage, to communicate and update or status that is then in turn translating into renewals that that closing that gap translates to renewals and and new business as well?
Brad Koontz 12:27
Yeah, that, that's great.
I mean these the use cases are fantastic.
It's so easy to see how you know how much value these things can add more companies.
How much faster they can get through things, less labor intensive, stuff like that.
But when we look at all of these, I would just call them individual improved workloads.
I would like to get both the else thoughts about this.
Talk to me a little bit about how we make sure that those things are aligned to their overall business strategy.
How do we untap that strategy and get it aligned with the gen AI road map for individual companies?
Jonathan Yundt 13:07
I personally I I think it's a supplement.
You think of it, it should be part and parcel of that overall strategy and thinking of it as another tool in the toolbox for that underwriter for that claims person for someone in operations for service, for marketing. How can I have that as part of an extension of them?
And there's, you know, plenty of articles out there where economists said that, you know, AI is not gonna take away your job. AI is gonna take away the job of someone who's not using AI.
So I think making sure that we're leveraging that and positioning it as a as a tool in in a way that they can consume that and shorten that gap is paramount.
Stuart Morris 13:51
And that's one of the pillars of our responsible AI strategy that we keep making clear to our users is that the output of these models, whether it's a synthesis or it's a content generation that they're doing for you, you know, we don't want these to go directly out the door.
We want them to assist a human who's already performing that job.
So any way that we can help expedite these processes, we we really want to make sure there's always that human in the loop, you know, verifying the content is correct and accurate.
And we're using that AI in a responsible way.
Jonathan Yundt 14:19
It's that 6 per sixth person off the bench, Brad, if you think about it.
It's part of the team. You call in that Gen AI, that AIas part of and you know. Obviously again we're not looking, Microsoft, for example, doesn't call it autopilot.
It's copilot. It's someone that's sitting next to you to assist you to extend your reach.
You know, you think of how there's only so many hours in the day, you know?
Help me synthesize this information, generate this information.
I use it on a day to day basis for, you know client presentations for catching up on.
You know, I was on a plane for 8 hours and I caught up on emails, but by looking at Gen AI is synthesis of what I missed.
I I think that's where we're seeing a lot of a lot of consumption as well.
Brad Koontz 15:10
Awesome. Jonathan Stewart, thank you so much for your time today and spending time with us discussing generative AI in insurance.
Stuart Morris 15:20
Thank you, Brad.
Jonathan Yundt 15:20
Had a blast. As always. Thank you.
Brad Koontz 15:22
Thank you for listening.
You can subscribe to exchanges with Hitachi Solutions wherever you get your favorite podcast.
You can find our entire episode library on our website at global.hitachisolutions.com. Thanks.