Sentry Transformation Saves ‘Tens of Thousands of Hours’ with AI in Claims

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

It would be hard to find a better testimonial to the power of artificial intelligence than what you’ll get when you speak to people overseeing an AI and technology transformation at Sentry Insurance.

The Stevens Point, Wisconsin-based mutual carrier insures 28,000 businesses of all sizes throughout the U.S. The company has taken advantage of the emergence of powerful new tools, like the large language model Claude from Anthropic for claim summarization, as well as new technologies to streamline document management.

Jim Frank, chief claims and information technology officer for Sentry, is overseeing these changes with much optimistic anticipation, and a large measure of satisfaction, so far. For example, he estimates that using AI for claim summarization alone is saving “tens of thousands of hours” annually.

Claims Journal spoke about the transformation with Frank, who has focused his last two years on leading Sentry’s transition away from legacy technology to cloud-based services and capabilities. Sentry decommissioned its mainframe in 2022, and Frank has been actively engaged in the development of Sentry’s IT and data assets. He has served as Sentry’s chief technology officer since 2012. He added the title of chief claims officer in 2023.

The conversation has been edited for brevity and clarity.

Claims Journal: What new technology have you recently integrated into your claims department/processes?

Frank: In 2024 we did implement artificial intelligence within our claim system specifically to generate claim summaries off of the file notes and the diary entries that that our claim reps do. So, as you can imagine, a claim can be over a long period of time, lots of evolving facts, and what we’re using artificial intelligence to do is create a synopsis of the claim, the facts of the loss, everything. So, as you transfer or move claims around, you’ve got an AI-generated summary in place.

There was a second initiative, which was really a total revamping and reengineering of our personal auto claims process and technology. We delivered a lot of capabilities there. That was the largest kind of lift and IT project that we did in 2023 going into 2024. We also have some AI things that are on the horizon that we’re currently working on related to document management and document intake.

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

Frank: In commercial lines claims can be long-tail, meaning they take a long period of time to manage and work to closure. There can be a lot of conversations, a lot of dialogue, a lot of back and forth and a lot of emerging of facts, and then you want to exchange or transfer those claims between people. The amount of time to uptake and to become familiar with the claim can take a while; that can be time consuming. Now, with artificial intelligence, we can create a summary that makes the hand-over, the transfer of the claim, very easy, very clear in just a couple of paragraphs. You can understand what’s going on with the claim and that prevents manual time that the claim reps had to write these summaries in the past and really kind of allows somebody to get up to speed very, very quickly.

We’ve measured a little bit of that and we are saving tens of thousands of hours per year by not having folks have to manually write these claims summaries. So, that’s why we’re implementing it and sort of the problem that it solves. Moving forward, we’re using artificial intelligence to scan documents and then extract information. Claim systems work very well with discrete pieces of data, dates, dollar amounts, facts and figures, and a lot of times that information is kind of stored or contained within a 50- or 60-page document we may receive, and what artificial intelligence is very good at is taking large unstructured data and turning in it into discrete, discoverable elements of data. And so we’re using AI in a number of places to extract information out of documents, put it in our system and then help guide claim reps in evaluating the document in its contents.

CJ: How is it working?

Frank: On our claim summaries it’s working very well. I would caution or advise CIO’s and chief claims officers that are beginning to look at generative artificial intelligence and large language models that you need to test very carefully. The results can be somewhat easily misleading a lot of times when you build a system. It used to be like grading a multiple-choice test, you know: Is it A? Is it B? But with AI, you’re really testing an essay: What’s the quality of the content? What’s the is it the right tone of voice? Is it the right generative output? And so we’ve really had to kind of scrutinize not just what it creates but really the true quality of the claim summaries. So, we’ve worked hard to refine it to make sure that it’s accurate and it’s precise. But it’s working very well.

CJ: How long has it been up and running?

Frank: On the claims summaries, those have been up and running since last summer of 2024. It was not a long or difficult implementation. You’re time-to-market on artificial intelligence solutions can be very quick because the technology is so powerful and so easy to use. However, we were very careful on the burn-in or the testing period, again making sure that we had the accuracy, the validations, the quality controls on it. So the time from the idea to our first implementation, you’re really talking days and weeks, and then we really let it mature harden the output of it over the better part of a quarter plus.

CJ: How do employees feel about it?

Frank: We engaged a lot of employees in the testing of the claims summary and the outputs within the application. We gave them some feedback buttons, kind of on: ‘How is this?’ ‘How do you like this output?’ So, we wanted a lot of user feedback to really validate, that we were getting human-like output from the machines as they’re writing these summaries. I think employees enjoyed that process and then also given the time savings that we’re generating from this solution, they seem to appreciate it that they don’t have to manually write these summaries and transcripts anymore. The machine can do it for them.

CJ: How do you feel about it?

Frank: We’re satisfied with the solution. We like that we’ve got AI in production, that’s a win for the organization. But also, I would say what we’ve done is really an entry and just the beginning of where we think we can harness artificial intelligence. We have more work to do, more benefits to realize, more use-cases to explore, so I don’t want to be overly celebratory with this. This is great, we like the solution this is proving the benefits.

But now the horizon is on more use-cases, more data extraction and utilization, and then more really harnessing this within the claims applications to bring more value. So a great win for our organization, a great win for our team; lots more work to do to expand upon what we’ve done already.

CJ: How will you measure the success of this technology? (What are the metrics?)

Frank: We had done some estimates at the beginning of the of the project of how much time claim reps were spending writing these summaries themselves and putting this together. And so you take the number of claims that require a summary, you know the number of times we do that in a year, and you estimate 30 to 45 minutes per summary that we have to write, and very quickly you add up to thousands of hours a year that our team is spending doing that. We’ve eliminated that step. You don’t need to write that summary anymore for this output.

We’re not specifically tracking ‘Well, what are you doing with the extra time.’ We’ve just given that back to the claim reps. We’ve eliminated thousands of hours from the claims process, and we can now move on to more value-add steps in terms of contacting customers, assessing liability and compensability, really the value-add steps of the claims process and not some of the more clerical or administrative tasks. That’s where artificial intelligence can help you.