To get the most out of using
Nationwide's two-pronged approach to Gen AI
September 9, 2024 4:40 PM
Laura Sauer
What applications does Nationwide find useful to harness Gen AI capabilities?
We are just launching our deep understanding of all the third-party Gen AI tools. All the big providers are embedding their own Gen AI tools in the different suites. That includes Salesforce, ServiceNow, Workday or any of the Oracle SAP products.
Which of these are we comfortable with our associates using? We lose some of that control over understanding exactly how it was built. What is it looking for? How is it coming up with the answers that it's giving back? Which of those are we going to be comfortable exposing to our associates to be able to use, and which we might not be? There's also a cost related to all of them. No one's giving you all this new functionality for free.
If our associates could get a huge amount of productivity or get hours back in their day so that they could focus on something else, then we really highly want to consider that. But we need to see that cost benefit, because some came out of the gate pretty cost prohibitive. We're a company of 24,000 and it's usually a per person cost. We're constantly trying to balance the financial aspect with the productivity aspect and the benefits to our associates. But it's a tough one, because everybody is now suddenly coming out with their latest Gen AI offering in their products.
Which of these are we comfortable with our associates using? We lose some of that control over understanding exactly how it was built. What is it looking for? How is it coming up with the answers that it's giving back? Which of those are we going to be comfortable exposing to our associates to be able to use, and which we might not be? There's also a cost related to all of them. No one's giving you all this new functionality for free.
If our associates could get a huge amount of productivity or get hours back in their day so that they could focus on something else, then we really highly want to consider that. But we need to see that cost benefit, because some came out of the gate pretty cost prohibitive. We're a company of 24,000 and it's usually a per person cost. We're constantly trying to balance the financial aspect with the productivity aspect and the benefits to our associates. But it's a tough one, because everybody is now suddenly coming out with their latest Gen AI offering in their products.
Can these Gen AI tool providers reduce the cost so it’s feasible for a carrier like Nationwide?
What we're looking at is not only the cost, but that value. We can create a similar tool on our own. It goes back to feeling confident that it's utilizing the right data and coming up with answers that we can validate are accurate. How is it so unique that I couldn't go build it myself? What makes it special enough that it's going to entice me to want to really use it? I believe that the cost will come down as more ideas keep coming out.
How does “Chat With Your Data” work?
It allows any associate to access a safe, walled off environment. It allows you to upload your own documents onto the secure site. You can ask it very simply, hey, can you summarize this document for me? Or can you compare these two items for me? And it's been really successful so far. In the first two days, we had 1,600 unique users within Nationwide try it.
We're trying to understand what they're using it for, get people excited about the possibility and then provide feedback on whether they got the right answer or what they were looking for. That's one really solid example that came out of listening to all the different business units for what types of things could be valuable for them.
We hope it becomes like an individual assistant. For example, in my role as a CTO, I'm training that tool to help me in my specific job. Right now, it's very generic, and it just depends on what documents you upload and what questions you're going to ask. The goal is to have your own virtual assistant that can help you with your specific job at Nationwide.
We're trying to understand what they're using it for, get people excited about the possibility and then provide feedback on whether they got the right answer or what they were looking for. That's one really solid example that came out of listening to all the different business units for what types of things could be valuable for them.
We hope it becomes like an individual assistant. For example, in my role as a CTO, I'm training that tool to help me in my specific job. Right now, it's very generic, and it just depends on what documents you upload and what questions you're going to ask. The goal is to have your own virtual assistant that can help you with your specific job at Nationwide.
What compliance items has the Red Team identified?
We have quite a bit going on in that space, the legal side. They have a list of 200 different types of potential legal and regulations that are out there across different areas. It starts with California and its Privacy Act. Oregon looks at it a little differently than California. The team understands what those different consumer privacy acts entail that impact anything AI or generative AI, then feeds that back to us.
The biggest achievement from the Red Team that we're really excited about is creating Nationwide's first set of ethical guiding principles for generative AI. It was rolled out to the entire enterprise to say, these are the expectations of the company about the ethical use of the tools, of the information. It really set a foundation for how we think about and how we expect our associates to work within this new technology environment that we're in.
We cracked a nut on getting in front of this ethical side of what to do with Gen AI.
The biggest achievement from the Red Team that we're really excited about is creating Nationwide's first set of ethical guiding principles for generative AI. It was rolled out to the entire enterprise to say, these are the expectations of the company about the ethical use of the tools, of the information. It really set a foundation for how we think about and how we expect our associates to work within this new technology environment that we're in.
We cracked a nut on getting in front of this ethical side of what to do with Gen AI.
Why are data risks a key part of Gen AI ethics considerations?
Data risk is Nationwide's second highest risk. Cyber is first, which is probably the same for every company out there. Data risk has multiple factors. We think about the quality of the data, the sufficiency of the data, the retention of the data, the privacy protection and ethical use of the data. These are all evaluated and tracked. You want to get to a certain point to feel like you're completely secure in that risk, that you have very little risk.
Data retention in technology has always been a concern. There's rules that certain types of data have to be kept for 10 or seven years. When there weren't any rules, we might have kept data for way longer than we needed. It's important for us to think through all our data and whether we're retaining what we want. Once you let things into a model, they're accessible. We want to make sure that it's accurate and it's the right data to share to get the right answers to questions and prompts.
That's given us a new way to think about data risk. Let's make sure that the data we've kept that is going to become part of these models is really what we would want to be there, not just because we had to keep it, but because it's the right information to be keeping and storing and then eventually building into these different models.
Data retention in technology has always been a concern. There's rules that certain types of data have to be kept for 10 or seven years. When there weren't any rules, we might have kept data for way longer than we needed. It's important for us to think through all our data and whether we're retaining what we want. Once you let things into a model, they're accessible. We want to make sure that it's accurate and it's the right data to share to get the right answers to questions and prompts.
That's given us a new way to think about data risk. Let's make sure that the data we've kept that is going to become part of these models is really what we would want to be there, not just because we had to keep it, but because it's the right information to be keeping and storing and then eventually building into these different models.
What is Nationwide’s Gen AI tool for claims agents? How was it built and how does it work?
It is an isolated pilot within one specific business unit, the claims area. The way that claims work with an insurance company is you could have to call in as a customer 10 times. First it's 'Hey, here's my initial accident, and then here's the update on what's happening with my car repair place, or whatever the next thing is.' Every time you call, the claims agent that you get types notes into a claims log.
Depending on what the claim is, that can be an extremely long series of notes and notes and notes. We noticed that a customer having one of those long, extended claims would have to call in, and then they'd have to summarize, because the new person they get on the phone might not be the same person they talked to before. That's a frustrating experience for customers. They say, 'Don't you know what is going on with my claim?'
The first thing our Gen AI claims summary tool does is summarize all the claims log notes into a nice, short paragraph that can be read super quickly by the claims agent that answers the call, instead of needing to ask the customer to repeat everything. It's a smaller, simple use case of generative AI functionality that's built into a lot of these tools, but has been a great opportunity for productivity and allows our customers to have a better experience when they call in.
Depending on what the claim is, that can be an extremely long series of notes and notes and notes. We noticed that a customer having one of those long, extended claims would have to call in, and then they'd have to summarize, because the new person they get on the phone might not be the same person they talked to before. That's a frustrating experience for customers. They say, 'Don't you know what is going on with my claim?'
The first thing our Gen AI claims summary tool does is summarize all the claims log notes into a nice, short paragraph that can be read super quickly by the claims agent that answers the call, instead of needing to ask the customer to repeat everything. It's a smaller, simple use case of generative AI functionality that's built into a lot of these tools, but has been a great opportunity for productivity and allows our customers to have a better experience when they call in.
What do you look for from a Gen AI insurtech startup seeking clients and users?
It's probably just 'show me something different.' There's a lot of similar ones out there. If you're really going to get a company like Nationwide, or frankly, any company, to bite on something brand new, it must have a really unique spin. If it's just going to summarize a bunch of data, we can do that on our own. If it's just going to compare two documents, I can do that on my own. It's got to be something unique [and] easy. That's the whole benefit of Gen AI. Making the process of getting it established, set up and production ready, as easy as possible, is going to be the big key.