Allianz Built An AI Agent to Train Claims Professionals in Virtual Reality

This article is part of an on-going series about technological and AI-related developments and advancements in claims processes and claims departments.View Series

One thing that experienced adjusters will probably tell you about learning to be a better claims handler is to get more—well, experience. Years and years of it.

Multinational financial services and insurance giant Allianz is looking for a way to accelerate that experience-getting process for claims novices and experienced professionals.

At its extensive North America operations, the company is experimenting with an artificial intelligence agent in virtual reality to speed up and improve a process that has historically entailed a training stint followed by years of on-the-job learning.

To get a good handle what Allianz is doing with AI, Claims Journal spoke with Melissa Hill, regional head of claims for Allianz Commercial in North America.

Hill has been in the claims field for more than 25 years. She joined Allianz from Sompo, where she was strategic claims transformation lead. She previously held senior claims positions at AIG, Hiscox USA and Kemper Insurance.

The conversation has been edited for brevity and clarity.

Claims Journal: What new technology are you integrating into your claims department/processes?

Hill: In claims, we’ve got an aging workforce and an increasing talent gap. I’ve been in the industry for about 25-plus years. I had the benefit of going through an extremely robust training program; it was about a yearlong training program. And then the industry, about 10 to 15 years ago, changed course. They hired young attorneys coming out of law school and then they pretty much just scrapped all of the training programs.

Fast forward to today: It’s been very difficult to hire college graduates, so as we have been collectively thinking through this. We’re trying to figure out ‘How do we remain competitive, resilient, reimagined? How do we train the next generation?’

One of the items that we’ve been throwing around is what I would call an AI agent, gaming system-type virtual reality training program. It’s a tool. Think about it like a gaming system where you put on the goggles, and you go into a virtual reality and you’re playing a video game. How do you turn that into an actual training program with situational training and analysis of an actual claim? We have been attempting to test and build and trying to reimagine how do we take a claims training program that got thrown away 30 years ago and reinvent that in the world today? We have all the available technology, so that’s where we’ve been heading.

Claims Journal: Why are you implementing this technology? What problem is it solving or will it solve?

Hill: We really want to attract the right college graduates and if you think about commercial risk, commercial policies have become very complex over the years. Your typical commercial policy can be anywhere from, I’d say 300 pages to 500 pages. That’s any typical commercial policy. You’re not going to hand a 300-page policy to a college grad and tell them to read it and figure out what that means. And in order for them to understand that commercial risk and what does insurance provide, you got to come up with a different solution.

So, what we’re testing is a gaming system, AI agent in virtual reality. So, if I’m the claims adjuster, I’m going to put on these goggles, I’m going to go into this virtual reality, I’m going to have an AI agent and then I’m going to have a risk manager. And we’re going to give them a claim scenario. As they go into that claim scenario, the risk manager is speaking to them, they are responding to the risk manager—it could be a problem, it could be an easy situation, a difficult situation. The AI agent is giving them feedback the entire time that they’re having that situation with the individual, but they can also stop it and ask questions: ‘What did I do wrong? What did I do right?’ (The AI agent is) telling them, ‘Your voice was too high, you were too demanding, you were rude or you were positive, you could have done this better.’ And what you’re doing is giving them that situational analysis and experience that they would not otherwise get.

Say they start handling claims, it might take them five to 10 years to get into a situation like that. Now, I’m giving it to them in the first six months of the job in a virtual setting that is a safe space, and they can get feedback and they’re actually understanding how the actual claim handling works through a much different process in this virtual reality from cradle-to-grave of a claim handling process. And you’re giving them different situations throughout that they would not otherwise get.

Claims Journal: How is it working?

Hill: We’ve been testing it and we’re using very experienced claim handlers and leaders to test it. When you go into it, you’re given a scenario in advance, you’re reading the information, you’re having these conversations and you would think it would be easy because we’re experienced claim handlers, but it’s not as easy as you think. As you’re getting pretty good feedback, (the AI agent is) reading the inflection in your voice, or (telling you) ‘You could have said this.’

And I think the interesting piece is the AI agent has the policy, they have regulatory information, they have the client’s portfolio, they know the client’s business. We are having the conversation with the risk manager—if we said something incorrect or we didn’t fully engage or state the process correctly, it’s giving us feedback to tell us that we should have said it this way, we should have made sure we read the policy on page 27.

So, it’s actually teaching you along the way how you should have had the conversation or that ‘You should have prepared better,’ and how you (should) actually approach the conversation. The agent’s getting smarter each conversation, each training and it’s very specific to a claim.

And what I like about it is you can actually plug in any type of claim situation. You could start using it for a live claim and you could practice before you actually have a conversation with a customer…it could be used as a training tool for actual claims in the process, to walk you through it. So far, it’s been a pretty robust tool and it’s taken off, and it hasn’t taken us very long to train the agent, which was very surprising to a lot of us that it happened pretty fast.

Claims Journal: How long has it been up and running?

Hill: It has been up and running for about three or four months in testing now. It took us a couple months to do the training and build the data dictionary and put all of the information in.

Claims Journal: How do employees feel about it?

Hill: The employees who are testing it, we think it’s pretty valuable. We’re thinking about all the other things we can use it for outside of training, but obviously we’re going to use it for training to start with.

Claims Journal: How do you feel about it?

Hill: I think it’s fantastic. I’ve been early adopter of AI and technology and I’m very passionate about training because I went through a very robust training program. I think it’s very valuable because that’s what kept me in the business for so long, is that I got really good training up front. It just gives you a lot more confidence when you feel like you know what you’re doing and you’re getting the appropriate training in the process. It helps you stay in the business, and I feel like it’s going to retain the talent.

You’re also bringing in college students who already know a lot of this technology, so we’ve got to give them the technology that already exists. Let’s build on what they already know. You bring them in, they’re going to teach us a lot more, and we can use this (to figure out) how we develop it further in managing claims. This is a people business, so this helps us communicate and accelerate the experience that we need to put into our talent. We’ve got to give them that experience, because they’re not going to wait 30 years to get the experience. We need to give it to them quicker and faster; we got to give them the knowledge to gain that experience.

Claims Journal: How will you measure the success of this technology? What are the metrics?

Hill: We have a graduate program here at Allianz, which is a fantastic program. We’re going to have some graduates that are coming in in 2026 who are going to be graduating out of college. We’re bringing some into claims; it will be the first time in many years that we are doing this for claims because it requires a license, so we’re going to put them through the licensing process. Part of that is they’re going to be training in this virtual environment and they’re going to help us build out this particular program and this whole training process.

They’re going to be part of this as new trainees and then after that yearlong program, they will go into handling claims. This is going to be (how we measure) our success. How does it work and how well do they come out of it? can they pass their licensing? Can they handle claims? Can they talk to customers? What can they do? This is going to be our true measurement, and how well they actually do in this graduate program. And if it works, we’re going to continue to do it and then keep building that bench for them with the graduate program.