How Can AI Help With the Insurance Talent Shortage? (2026)

The insurance talent gap and where expert time goes

Jill Zakarian, BevayaThrilled to be joined by our co-founder and CEO Chaz Perera. So Chaz will take us through how underwriting and claims teams are using AI to manage growing workloads, scale their knowledge, and make better use of their expert time. So without further ado, I'm happy to pass things over to you, Chaz.

Chaz Perera, BevayaAwesome. Thanks, Jill. You know, finding experienced underwriters and claims professionals has been hard for a really long time and it's getting harder and the work, frankly, keeps growing. Maybe you have an underwriter seat that's been open for a few months.

Maybe your most experienced people are carrying more files than they have in prior years. Either way, you're probably feeling some version of that right now. Today we'll cover where the talent gap stands, where expert time goes, how AI and your experts divide the work, how teams keep their experts sorry, how teams keep what their experts know and how to plan capacity as your volumes grow. We'll close with some results and how to start.

So about 400,000 insurance professionals are expected to leave the industry by the end of the year. Much of that is due to retirement. And you've probably seen it on your own teams. Many of the industry's most experienced people are reaching the end of their careers at the same time.

Fewer people, by the way, are coming in to replace them. And even when you do hire, it takes years to build the judgment a senior underwriter or a senior adjuster brings. Meanwhile, 73% of insurers say they're most likely to hire experienced talent. Everyone is competing for the same small pool.

Half of insurers plan to grow their teams this year and 43% plan to hold headcounts steady. The most, by the way, in 15 years. So, most teams are either hiring into a tight market or asking the people they already have to take on more work. Volumes keep climbing and the people with the experience to handle it are getting harder to find. The time your experts do have, by the way, is really worth protecting.

Judgment work vs. setup work

Chaz Perera, BevayaSo, where do these hours go today? Map out a senior underwriter or adjuster's week and split it into two components. On one side is the judgment work things like coverage calls, pricing, considering sort of the idiosyncrasies of complex claims. These are the decisions only your most experienced people can make and the work you've hired them to actually do.

On the other side is the setup work. Picture a submission that shows up as an email with half a dozen attachments. The ACORD, a few loss runs, a schedule of some type. None of it in the format your systems want.

Or maybe it's a claim file where documents arrive from email, the mail room, and brokers all mixed together. Somebody has to open all of that, sort it, and key in data that's already written on that page into your systems. It happens, frankly, before any of the real work starts on every file every single day. And it doesn't take 15 years of experience to do that.

So, as volumes grow, the setup work grows with it. And it takes a bigger share of an expert's week every single year. When a senior person leaves, their setup work stays behind. It lands on whoever's left, and that extra load can push the next person out the door.

The good news is that this part you can change. You don't have to wait for the labor market to loosen up.

What AI can take on in underwriting and claims

Chaz Perera, BevayaOkay, that split is where AI comes in. On the left is the setup work AI can take on today. In underwriting and in claims. In underwriting, that's submission intake and clearance.

It's triage and routing, loss runs, prepping the underwriting and prepping the underwriting file. It also covers things like renewals and comparing one policy to another or one quote to another. In claims, it's first notice of loss setup. It's indexing all the claim documentation that's coming in.

It's making an initial assessment around coverage. It also covers pulling details out of legal demands and medical bills or processing vendors invoices or frankly writing claim summaries so you can hand a file from one adjuster to another. So on the right is what stays with your experts. In underwriting that's the decision to accept or decline a risk, the pricing judgment and the broker negotiations.

In claims it's the coverage call ultimately. It's settlement. It's the litigation strategy and the decision to refer a claim for fraud. Let's take claims indexing.

The AI reads every page that comes in and sorts it by document type and pulls the details that tie it to the right claim. Then it routes the document where it needs to go. The adjuster opens a file that's ready to work instead of a pile that they have to physically sort first. Now, let's think about it from an underwriting file prep point of view.

AI can build the file. It can check the data. It can flag gaps before you get to pricing. The underwriter still has to make the pricing call.

They just start from a more complete place instead of a stack of documents and attachments they need to act on. The goal is that the work in front of your experts is the work that actually needs their expertise. Picture a Monday morning. Any Monday morning, an adjuster opens their queue and the routine files are already set up and ready.

What's left is the work that calls for an adjuster, the disputed coverage, the pattern that looks a little bit off or the claimant who needs a real conversation to get to a good outcome for both parties. That's a better job and a better use of someone you worked hard to hire.

How teams keep their experts' knowledge

Chaz Perera, BevayaSo freeing up time solves one part of the talent problem. The other part is what happens to everything your experts know when they leave. Most of that knowledge lives in our heads. It gets passed along from conversation to conversation, often one conversation at a time.

And the people who hold the most of it are usually the busiest people on your team. There are four ways AI can help that knowledge stay with the team. Corrections from your reviews retrain the AI models. Your rules and guidelines guide the work.

Your experts decide what needs review. And when someone new does a review, they get the context they need. So let's go through each of these one at a time. When the agent isn't confident about a value, it doesn't guess.

It sends that item to a reviewer with the source document right there in view source system too by the way. So the reviewer can check it quickly instead of hunting through a 40-page file. The reviewer confirms it or fixes it and every fix feeds back into the model. That's a feedback loop.

Your reviewers don't do anything extra to teach it. Fixing the value is the teaching. Over time, that adds up. The judgment your senior people use today doesn't stay locked in their heads.

It builds up in the system file after file. So when someone retires or moves to another carrier, you lose less of what they knew.

Rules, confidence scores and human review

Chaz Perera, BevayaThe second is your rules and your guidelines. Think about what your most experienced people carry around with them. Your appetite, your underwriting guidelines, the way your team handles a legal demand. You know, teams write that knowledge into rules and the AI checks every file against them.

If a submission falls outside your appetite, it gets flagged before an underwriter spends any time on it. On the claim side, a playbook for legal demands means every demand gets read the same way and the time-sensitive requests and the deadlines get pulled up and out upfront. When a rule changes, the changes also apply right away. This helps a stretch team in three ways.

The judgment stays when the expert leaves. New hires work from the same rules as everyone else from the day they show up. And you spend less time training new staff on the basics because the basics are built into the process right from the get-go. The third way is deciding what needs review.

Every value the AI pulls gets a confidence score and your team sets the line the threshold around these confidence scores. Above it, the AI moves the work forward. Below it, a person takes a look. You set that line field by field.

It can sit higher where the stakes are also higher and lower where a miss is easy to catch and doesn't really change the outcome. Your team decides where based on how you run your business. The scores mean what they say. When the model reports 90% confident, it's right about 90% of the time.

That's what makes the line worth trusting and what makes it safe to hand setup work over to AI in the first place for your experts. By the way, review time goes to the file that needs them and the rest move on their own because senior people set the judgment line, the threshold. Their judgment also keeps working after they move on. The fourth way is for your newer people.

When someone new reviews an item, the system gives them the context a senior expert would usually apply so that they can make the call without someone having to look over their shoulder. They see three things. The exact spot on the source document or system or page behind the extraction or the inference. A confidence score so they know where to look closely and a short summary of what the item is and what needs attention before they have to open anything and before they have to figure things out on their own.

Think about a new adjuster's first few months. A lot of that time goes to asking a senior colleague where to look or what really matters. So with context built into every review, more of these answers are right in front of them from the get-go and it helps shape their education a lot faster as well. So teams can expect faster ramp up for newer staff, less time spent shadowing senior people, which I think is really important and a clear sense of what to check first. So your senior people, frankly, get some of their own time back, too.

How to plan capacity as volume grows

Chaz Perera, BevayaNow once AI frees up hours, the next question is how your team will use that new time. You can work through it in four steps. First, count today's hours. Know how many expert hours each workflow takes before anything changes.

It doesn't need to be a perfect math. A rough weekly count by workflow is enough to start and you can tighten it later. Ask a few simple questions. How many submissions or claims come in each week?

How long does each one take to set up? How, or rather who, does that setup work today? You may find your most senior people spend more of their week on it than anyone actually realized. Second, once AI is set up and running, track two numbers.

One is the expert hours freed up each week. The other is how much work AI handles start to finish without a person having to touch it. We call that the no touch rate. Compare both to your starting count and you'll see the shift in your own numbers.

Third, decide where the extra hours go. Some teams clear the backlog of complex claims. Some take on more submission volume without having to add people. Some finally make time to coach new staff.

Others give underwriters more time with brokers on their biggest accounts. Others let adjusters spend more time talking with claimants or put senior people on writing down the guidelines and playbooks the rest of the team really needs to work effectively. Any of those can work as long as someone decides so the time doesn't drift back into emails or whatever request is loudest that week. Fourth, share the results with leadership.

Operations leaders want to see the hours saved. Finance leaders want to see growth without new hires. Those are often two different people and you'll want both on board. Leadership tends to hear we freed up time.

It's a nice extra. They hear we freed up this many hours and here's where that's going. Sounds like a plan. The second version is the one that gets continued support for more AI and for using it past the simple workflows.

Results from insurance carriers

Chaz Perera, BevayaNow, insurance teams already working this way are getting real hours and capacity back. For example, a regional carrier lifted its claims team's capacity by 85% by automating invoice processing. A top 20 P&C insurance carrier cut ACORD processing from 10 minutes to 2 minutes. A US property casualty carrier doubled its submission intake without adding any new headcount.

And another US P&C carrier saves more than 36,000 hours a year just on FNOL processing alone. That's capacity, speed, volume, and hours back across claims and underwriting. In each case, the same people spend more of their day on the work that actually needs their judgment. So these teams got more out of their experienced people that they already had and they were able to measure it and give quality perspectives around those challenges.

How to start: pick a team and a workflow

Chaz Perera, BevayaSo if you want to try this yourselves, where do you start? Pick one workflow that's high volume and well understood. Intake and indexing are common first choices because the work is repetitive and the results are easy to see. Well understood matters as much as high volume.

Pick a workflow where your team already agrees on what a good file looks like. That makes it easy to set a starting count and easy to tell when the AI agent gets it right. Save the messiest, most debated workflows for later. Once you have a wind behind you, then you can go fast.

Start with a single team that's open to it and let them review and correct the AI's work as they go. That team shows the rest of the organization what works and often becomes its best advocate. Open doesn't mean everyone has to be a believer on day one. Include at least one experienced person who'll push on it or push against it. So to speak.

If the AI holds up to their hardest files, their hardest judgment, their hardest resistance, the rest of the team will trust it faster and their corrections are exactly what your feedback loop needs anyway. Then check the results at 90 days. Compare expert hours to the count you took at the start.

Look at how much work moved through without a person touching it and check whether the extra hours went where you hoped it would go. Then make a clear call. Expand to the next workflow. Adjust what isn't working or just stop.

Any of those is a fine outcome. As long as it's a deliberate decision, that's enough to make the case without betting the whole operation on day one.

Key takeaways

Chaz Perera, BevayaThe scarce resource here is your experts' time. AI can take on the setup work so that time goes back to judgment. And the way your team reviews and guides the AI builds what they know into how the work gets done. You don't have to win the hiring market this year to make progress.

You need more room from your experts. Thanks everyone for joining us today. Jill, I'm going to hand it back to you for any Q&A.

Jill Zakarian, BevayaThank you, Chaz. We hope everyone's leaving with a clear picture of how to give your experts their time back and keep their knowledge with your team even as volume grows. And before we hop into our Q&A, I'd like to invite you all to check out the many other past webinars available on demand. You'll find a link to our video library in today's presentation slides.

Q&A: finding time lost to intake and rekeying

Jill Zakarian, BevayaAll right, we're going to move into our Q&A. First question that I see in our queue is, we suspect our underwriters lose a lot of time to intake and rekeying, but honestly don't have a hard number on it. How do teams figure that out without launching some giant time study project?

Chaz Perera, BevayaI'm personally not a big believer in giant time study projects. You know, I would argue you don't need something overly formal. Especially if it's your first time, I would say start rough. Pick one workflow.

And you know track it for a week I suppose. Have a handful of people ballpark how much of their day goes to you know activity one versus activity two versus activity three in that process. Understand for each of those activities where is real judgment being utilized. You know and then I think you might be surprised at where the real numbers land or the estimates land.

And I also in my experience have found that the experts in your organization don't even need a week to tell you this stuff. If you sit with them for an hour or two, they'll be very candid about where their time gets misused. Yeah, I would say those are sort of the things I would say just start rough. You know, I suppose one other thing I'd add here is from a, in our experience, in commercial underwriting, for example, 40% of an underwriter's time is focused on manual reading, keying, reviewing documentation, etc.

That ultimately gets to making a judgment call, but is not the thing that is where their time is best used because you really want them focused on is this the right risk at the right price? How do I get the broker on the same page?

Q&A: what to capture before an expert retires

Jill Zakarian, BevayaAll right, awesome. Next question is, one of our most experienced adjusters retires next year. What can we realistically hold on to before she leaves?

Chaz Perera, BevayaWell, you know, the year's around the corner here, so work fast. You know, this might sound a little like the answer to the first question, which is you just got to sit down with this individual and spend a couple hours trying to understand, you know, what they do and try to capture their calls and you know, when they start their day, what's the first thing they do? What's the first thing they check? When they think about red flags, what are the red flags that they're watching for?

You know, typically when a hiccup happens or there's an oddity, you know, what were the things that they were expecting versus, you know, what they're seeing. Those are all sort of useful hints that will shape you know how you think about where rules and guidelines need to be enhanced. Ultimately if you even you know with the power of AI today you could take that brain dump so to speak give it to AI and say help shape this into a series of guidelines and rules for me that then I can instantiate in an underlying process in a consistent repeatable way.

Once you've done that, get the AI to sort of work on that process and leverage those rules and clearly that you know this adjuster has told you this is where I'd like to be. I'd like to get involved and so the AI needs to make sure that that adjuster gets involved in those steps. And you know, ultimately the way to test that really is to you know, whether it's in a sandbox environment or in out in the ether, so to speak, in the real world, because you've put the right thresholding in place to get to a human, put the hard files in front of the AI to see if the AI is adhering to it properly.

Q&A: how long AI takes to set up

Jill Zakarian, BevayaNice. All right. Next question we have, Chaz, is our team is already stretched thin. How much are their time or sorry, how much other time does it take to get something like this up and running?

Chaz Perera, BevayaYou know, there's sort of like a hard ask in that question for your people because the people that are stretched thin are often the people who you need to rely on to get this information. So, it's a big ask up front. But you know if somebody knows the workflow really well and they really understand what those workarounds look like and you know how exceptions come to be. What choice do you have other than to spend a little bit of time with them trying to map those things out?

For context, when we're implementing our AI with a customer, we typically need to spend anywhere between 90 minutes to two hours with the insurance experts to just make sure that what we understand that process to be and how they conceptualize the process are similar. And we take the delta between that from that conversation and give that to our AI to say you know yes you understand this process you typically works like this but you need to factor in this delta for it to be an effective outcome for you know customer X. And then you get to production and you run it in production and you set the thresholding low and that expert who gave you that perspective has the opportunity to review those things in real time and make assessments or judgments about that shift.

So you know you're talking 90 minutes to two hours just to understand the process. Then you're talking two to six weeks to go from that 90 minute two-hour meeting to something that is production grade and up and running creating value with it with those experts on top of it. So, you know, relatively short amount of time with the right scaffolding. And, you know, that's what we would argue our technology, our AI is the real scaffolding to help you implement AI but scale it successfully.

You know, if you're trying to do it on your own, typically that you're looking at you know, a few months to a few to theoretically a few years depending on the complexity to get there. But, you know, it really comes down to expert time is not as extensive as you might imagine.

Q&A: how reviewer corrections train the AI

Jill Zakarian, BevayaAwesome. All right. So, I want to take one last question. Chaz, we had a lot of questions that came in.

I know we're kind of coming at time, but I think we have time for just one last one to squeeze in. So if reviewer corrections are feeding back into the model, how do we know it's learning the right things and not quietly picking up one reviewer's bad habit and baking it in across every file?

Chaz Perera, BevayaSo I think probably the most important thing to start with is the review process you have should be highly structured, right? You want to direct the way a human sees the problem and provides a response or an answer to solving that problem in a deterministic way. So it's not a hey I would do this and then I would do this and I would do this. No, it's more like you saw this, I saw this, I say this, right?

Much more structured. Then I think it's important to set a threshold these confidence thresholds that at a level that you're comfortable with initially. For some customers by the way that could be for us today some customers are like hey look my people are 90% right on this stuff. So I want the AI to be 90% right on this stuff.

And then everything you know below that threshold goes to a human to look at. And then what you end up having is over time as the AI sees these answers not just from one individual but a series of experts and that's important here a series of experts providing input as opposed to a single individual you get a mixture of experts to use an AI term in the wrong sense here so to speak but a mixture of experts providing the right input to shape the AI and the AI's knowledge and the AI's memory then you're getting to a good outcome. Very quickly.

Jill Zakarian, BevayaAll right, awesome. So, we are going to wrap up here. We appreciate you making time to join us today. I know there's a lot of questions that we didn't have time to cover, so we will answer them after the session.

Also, feel free to email us any questions you think of later at resources@bevaya.ai. And as a reminder, we'll share a recording of today's session by tomorrow afternoon. We hope to see you again soon and have a great rest of your day.