Plymouth Rock's Jamie Warner on Agentic AI and Data & Closing Remarks

Transcription:

Patricia L. Harman (00:14):

Welcome back to the Digital Insurance Virtual Summit, ATech AI and Insurance, the next wave of enterprise transformation. Now we're going to explore agentic AI and data, I'm sorry, goal-driven decision making and the impact. And joining us today is Jamie Warner, managing director of data science pricing at Plymouth Rock Assurance. It's so nice to see you again, Jamie. I'm really excited to have you here.

Jamie Warner (00:45):

Absolutely. I'm thrilled to be here. Thank you, Patty.

Patricia L. Harman (00:48):

So we are going to talk about all things agen AI and agentic AI is expected to significantly change how data is used, managed and valued, shifting it from being just like this static asset used for retrospective analysis to really a dynamic input for continuous real-time use. And so from your perspective, where does the insurance industry stand at present in terms of ag agentic AI and data?

Jamie Warner (01:24):

Absolutely. Well, I'm so excited to talk about this. I know I saw a stat, I think it was from Gartner recently where less than 1% of enterprise software applications use this right now, but they're expecting over 33% by 2028. So just explosive growth in this space. And it's a space where I'm really excited because we've talked in the past about insurance having a little bit of a data problem and a collective data problem. And this is really a space where we're going to have a lever with which to push ourselves into improving our data in ways that will help, not just with agentic ai, but regular AI data science analytics. And so I think the biggest challenge right now for insurers is that a lot of our systems still don't talk to each other and collecting the information we'll need to make these agents effective isn't always going to be super easy.

(02:14):

So for example, the historic framework for it is you take the data, you put it somewhere and you store it maybe for audit. And then in recent years we said, okay, well we want to take that and use it for analytics, but it doesn't necessarily have to be real time. It can be next day or whenever. When I think about some of the use cases for agentic ai, I always go to claims first because it's one of the places that's easiest to make test and learn cases. And so if we think about automated claim settlement, well all of a sudden you need those data flows to immediately flow to those tools. And so that's building a whole new type of infrastructure that isn't really our historic IT idea that it will flow through and continue to add more information as it goes. So I think we're at the very starting point, but I'm excited to see how well this pushes us into the next phase of data infrastructure and IT infrastructure in an insurance industry.

Patricia L. Harman (03:09):

That's a good lead into my next question because I want to know how should insurers prepare their data and systems to utilize agentic ai? I mean, I think about how it's in all of these different places and all of these different formats, and so what do they need to do now to kind of get their data ready so that they can actually utilize agentic ai?

Jamie Warner (03:32):

It's a great question and I think the first thing is to make the big problem into a small problem. So if we think about this task, it's immense. And if we think about the combined ratio and the percent we want to spend on it, that's not necessarily immense. So I think the best place is really to look at the list of use cases and say, which one is the best one for us to start with? Which one will make the most business impact for us? And start with the systems that are connected to that use case. So if we go back to the claims example I was using, you want to check your claim system and first with the vendor, is the vendor starting to do things that make data more easily collectible from their tool? What are the flows look like today and how easily can you get it?

(04:15):

And can you work on some temporary solutions? So just because you can't get it next minute, maybe it's okay if you get it next day for now while you're building out something and getting kind of the benefits of that adaptation. So I would really think business case first and not let's transform all of our it because a multi-year multi-million project, sometimes billion depending on the insurer, it's really what is the use case that's most significant for my company? Is it fraud? Is it claims? Is it something with our risk assessments? And then focusing in on that one and that data flow immediately.

Patricia L. Harman (04:50):

It's kind of like that old saying, how do you eat an elephant? It's one bite at a time. And it's the same thing with your data. You just can't approach everything at once. So I like how you said, prioritize what you want to do and go from there. Where do you see having this change and the use of ag agentic AI eventually having the biggest impact then?

Jamie Warner (05:12):

I love the potential impact with customers and getting tools, getting things into their hands faster. I think one of the hardest things we're coming into hurricane season, especially when we have these fluctuations in staff. So for call centers, like customer service and for claims, there's always in the time of most need going to be a fluctuation where we don't quite have the ability to cover that need because we're staffed for a certain level and then when extreme events happen, that's when we're needed the most and that's when we have the most strain. And so where I'm really excited is how can we lessen that strain so that the people in the loop, the human in the loop aspect can be doing the things that are the most critical for people to be working on. And kind of these tools can be taking on the tasks that are more obvious. And so I would say after a big storm or a hurricane, can they be auto adjusting things that can be auto adjusted? Can they be getting information out? Can they be doing risk assessments on images that are coming in from the customers? All those sorts of things can help us settle claims faster and get people back to where they need to be more quickly, especially in those more traumatic situations where there is pressure on the humans and they can't take it all on.

Patricia L. Harman (06:28):

What kind of timeline do you foresee for the industry? We both know that adapting new technology has not always been something that the industry has been very rapid about doing. So what kind of a timeline do you see for this adoption of agen ai?

Jamie Warner (06:46):

I've been really thrilled for what I've seen in the claims in the call center spaces. A lot of companies are just jumping right on this and obviously there's some holdouts, but we're seeing a lot of insurers say, let's test and learn it. Let's try it out. I think the space where we've got a really long haul is pricing, and that's just I think department of insurance. You think about the turnaround on a current pricing model, even if you could fluctuate it day to day, number one, you might not want to. And number two, the approval process is much longer than that. And so how do we measure if that's appropriate, how do we understand if it's accurate, all those sorts of things that the department is watching out for and how do we give the departments better tools to analyze the work that we're doing?

(07:34):

I think that part is a really long way off. And so that one, I feel like we've got a big barrier already, even to just large language models and other sorts of tools. I do think this will also help a lot with the image work that we do. So that's a place where I think we can be pretty rapid and we're already seeing the large language model work on image recognition and what's in the image, what information are we pulling out? So if you think about some of the stuff we've seen in auto, there are plenty of companies that say, send me a picture of your wrecked car and we can adjust it right away. Those sorts of use cases are perfect for this and I'm hoping it'll expand it exponentially. Those are the sorts of things that we don't necessarily need you to wait for someone to check because they're getting so good at that image detection.

Patricia L. Harman (08:23):

I think it also will help with even identifying fraudulent claims, again, because of the image recognition and how it can be used. And it'll recognize that, oh, somebody pulled this off of Facebook or they pulled this. I remember one adjuster saying to me, oh yeah, when I get a claim for lost watches, I go to the Rolex watch site to see what images they use there because sometimes people will use those as part of their claims. So I'm guessing that agentic AI will help to flag some of that too, kind of going forward.

Jamie Warner (08:57):

Absolutely. I think that's a brilliant case and I think that it really helps because fraud is such a huge problem in our industry and it's so hard to identify.

Patricia L. Harman (09:08):

True. What excites you the most about the possibilities of agen AI and data?

Jamie Warner (09:15):

Well, so I know you've heard this from me before Patty, but I just love the way we're going to be able to get data out of all of those things and insurance that we haven't been able to get data out before. So examples would be so much of our historic and even current documents, our PDFs, doctor's notes, images, and it's really hard to get usable data out of those things. And these tools have made it almost the easiest part of our process when previously it was the thing we were hiring the most PhDs and analysts and other things for, and we were doing all this manual labeling. And labeling is still a big part of that, but it is really going to open up our opportunity to be so much more effective with our models and measure their effectiveness. And that changes I always love as far as we can go from a proxy variable, like something where we don't have the real data, we're just trying to get as close as we can to the actual data. So the stuff that's in an image, the stuff that's in the doctor's notes, we're not trying to predict what you're telling us. We actually have it and we can really use it. So that is thrilling.

Patricia L. Harman (10:20):

I was interviewing two physicians and they were very excited about how they can use ai, and they said AI has the ability to see what the human eye cannot. And that goes to what you were just saying. And so in terms of using agen AI to collect this information and kind of help form ideas and say, Hey, this is what you need to look at, or maybe you need to consider this type of diagnosis, or maybe this could be, this could be what's causing your patient's problems, that just opens it up to a whole nother level of opportunities I think across a lot of different industries.

Jamie Warner (10:59):

Absolutely. And we've seen that in the medical field already pretty significant. I would say precursors are being discovered. So for example, they'd have a cancer drug where they notice it only has a 4% effective rate, so maybe historically that wouldn't go forward. Now they're seeing, well, almost everyone that was in that 4% had this same thing on their x-ray or image or type of genome sequence. And so we can also start to use that same technology to understand these homes all had a fire. What is it that drives that unique existence? And especially with life insurance, with home claims, with all of these spaces, our claims frequency is really low. So it's hard to say, what are the things that make this have a claim and this not? And this can really help drive the answer to that question, and that'll help us give people more effective pricing. It'll help us be more prepared for resilience and maybe even prevent some of those risks. So that place is really the limitless opportunity.

Patricia L. Harman (12:02):

I agree, especially the idea of being able to prevent risks or even to mitigate their impact. That will have huge impact, a huge effect both for the policy holders and for the carriers. So rather than human defined queries, AI agents proactively seek out what we were just talking about, new patterns. Yeah, I can't even get the word out. Anomalies. Yes. Getting to the end of the day correlations and it uses data for autonomous exploration and hypothesis generation. What kind of impact will this shift in how data is used have on the insurance industry broadly? Then

Jamie Warner (12:50):

I just love the ability for it to run iterations and then humans can come in the loop and say, oh, that's something I didn't notice. Because one of the things that is great about insurance is we have all these brilliant people that work in this industry. One of the things that's hard is once you're in insurance, a lot of times you're in it for life. So you're not necessarily stepping back and going, how much has the industry changed? And if we think about the change in the industry and the change in behavior even in the last five years, let's think about the pandemic. If you think about just home insurance, people suddenly have more dogs, suddenly they get delivered more packages, they're in their homes more. Those changes in trend are things you might not be thinking about. If you've been doing this for a really long time and you think you know what your population looks like, you think you know how they behaviorally act.

(13:37):

And so a lot of this modeling and looking for information that these models do isn't being done in this way where you're thinking with your preconceived biases and beliefs about how it should look. It's really just looking at what the data says. And so that can give us insight where people can start to question, okay, is this bias or belief correct? And if it's not, how are we going to change or adjust to make our pricing more effective, make our claims process make sense, whatever it is. So I'm excited that it'll give us that kind of extra information to really drive the change versus having to drive the change for how we think things operate.

Patricia L. Harman (14:14):

Yeah, there have been so many changes in the industry, and you were talking about people being in it for life as somebody who's covered the industry for 30 years, I totally agree. I think it's fascinating. And when I talk about insurance to people, they're like, oh, you must be new. I'm like, no, I've been doing this for a while. And they're surprised because I'm so excited about it and I am in a unique position. I get to see how the industry changes, how the risks change, how technology is applied and the difference that it's making both for the carriers, the agents and the brokers and for their policy holders. So it's really, for me, it's been a very interesting evolution to watch.

Jamie Warner (14:56):

Yeah, and it's funny you mentioned the brokers because it's a place that I don't talk about a lot because that's not as much in my personal space, but I think that the brokers is one of the coolest places to use this type of work. Thinking about policy documents and how hard they're to read and how difficult it is to manage the information and get that information to the customer, that's a place where this sort of stuff can be used today, not the regulated space and can really add a lot of value because people don't understand what's in their insurance policy. They don't understand what to ask for to get what they need. And especially when you're dealing with tons and tons of policies, you can't spend as much time with each of them to train them up on what is insurance. So I think that's a super exciting space.

Patricia L. Harman (15:43):

Imagine if policy holders use chat GPT or any sort of AI to say, explain my insurance policy to me. Imagine what a game changer that would be for them to help them understand.

Jamie Warner (15:58):

And if we can build agents that can control what it says back so that it's not just G hallucinates or gives them extra information or more generalized information, it's great. Because when we think about how we can bolt this into our systems, we can put our policy language in there, we can put their information in there and we can have it give them a personalized experience, which is going to be incredible because that's something that we always get as a complaint. I don't understand my policy, I don't understand what's in it. Well, let's make sure you do. And this is a tool to do that,

Patricia L. Harman (16:28):

Right? Totally. And again, it goes to that customer experience and just trying to make it as positive as possible for them because such a priority for all the carriers. So shifting gears a little bit today, data lakes and warehouses store data for potential use with agent ai. Data can be pulled, filtered, transformed, and evaluated for its relevance to achieve specific goal or goals. And the impact for insurance companies could really be huge across business lines and operational areas. In what areas in the insurance industry do you see this goal-oriented data usage having maybe the most profound effect?

Jamie Warner (17:14):

Yeah, I think so many spaces, but primarily one of the things that we talked about at the very beginning was how it's such a challenge to get our data in the right shape for these tools and suddenly these tools are able to help us with that. So if you think about if you have legacy code that you need converted, these tools can now give us part of the way or most of the way. If you think about whether your data is labeled correctly. I know for example, snowflake, which is a popular database tool, is adding in features that leverage the information within your data and help you kind of automate some of that governance, some of that information gathering some of that metadata gathering so that you can, instead of having a person have to go through and see, wait, what was this? How did we get it?

(18:00):

Where did it come from? How was it transformed? They can start to do all that for you, which really enables you to more rapidly adopt the rest of the tools more rapidly, be able to leverage AI on top of it. And so I think the current revolution is almost within it. Of all your programmers that are helping build these apps or tools, they can get a help from copilot, for example, which will help them pre-write their code. Again, we can't use it a hundred percent of the time. We do have to have a human loop to check it, but this really gets us a jump in productivity, a jump in the ability to understand our data and the patterns in our data and oh, this data year to year had a big shift. We don't have to manually do those calculations. We can use these tools to do that. So that is really making it a lot easier to transform and get through that whole process. And I almost feel bad for folks that started five or 10 years ago ahead of the curve because it's so much easier now to do a lot of that work

Patricia L. Harman (19:01):

And imagine what it'll be like just in another two or three years. I used to ask people questions about, well, where do you see the industry in the next three to five years? And I can't even use that timeframe because technology is changing so quickly and it is just having a profound effect in how businesses operate and how they service their clients and the products that they're developing and that they offer. It's just really, it's very interesting to see all of this unfold. So Jamie, you're a data scientist, so maybe you can help the audience understand how agentic AI will allow data flows to be redesigned in real time adapting to new sources or objectives without human intervention. How easy will this be to achieve?

Jamie Warner (19:57):

Yeah, I think it's a great question. Easy is kind of a relative term, so definitely easy to achieve. The question I think is a little harder is how effective will it be, how accurate it'll be, and how do we measure those things? What are the gates we put in place to make sure that it doesn't go off the rails? And this is an area that's not an insurance specific problem, which is why it's really fun because it's actually a global problem. And the top research minds right now are on the case of how do we measure these things? How do we stop them, how do we figure out where they're the most useful? And so we really get to borrow all of their research because we're doing a lot of that. And I think the other place that we really need to think in this space is what does the customer feedback loop look like?

(20:47):

So you've probably noticed on some websites when you get a bot answer, you get it faster and then it might say, how'd the bot do? And you might say, terrible. And so one of the things we have to think about is these things will be most effective if we have a way for people to opt out really quickly when they're not effective. And so that's something I know we've thought a lot about of hey, if it's getting you what it needs, great, have them keep clicking through. If not, reroute them quickly so that the person can ask. And also you can use voice analytics on that call with that person asking, and those can go back into the system and retrain for the next experience. So basically you just want to make sure, and this is a really new problem for insurers, that you're collecting data at each stage so that you can feed information back into it so that it can be better trained next time.

(21:34):

You don't want it to be this situation where it goes to the bot, it's not happy with the bot, and you don't know why because you can't take an action. So I like to think of it as that really, if you think back to high school statistics, you did the Y equals MX plus B, you want to know if I change X by this amount, how much is Y going to change? And what is my X? What's the lever I can pull to change people's impression of this or effectiveness? And so are you collecting the data to actually tell you that? Do you know where they click next? Do you know what they do next? These are all types of data. We haven't really tried that hard to collect historically. So it's all about that collection and then putting it in a place where people can actually access it to do the analysis or flow it back through a model so it automatically updates based on the new information that comes in each day.

Patricia L. Harman (22:22):

That just is stunning to think of how much more data and information that is, but then it'll be so valuable in terms of helping you solve those problems. I remember the first time I was on somebody's website and the bot popped up and I was just like, what is this? And I kept trying to close it and I didn't understand. And then now I'm so used to that happening and I'm encouraged to hear you say that if it answers the questions, great, we can keep doing it, but if it doesn't, we need to reroute them quickly. Because I think as customers and consumers, that is probably one of the most frustrating aspects for us is because I know it's like when you're on the phone and you're trying to get a real person and you're like, please just give me a real human being. All I want is a human being, please. Isn't there another live person there? It's like what's the magic word to get out of wherever it is I am and get to the place where I need to be? So how quickly could data ownership move away from a centralized role-based distributed usage role-based to distributed usage with access governed by agents needs, permissions and goals, it kind of shifts the focus. And who has access at that point?

Jamie Warner (23:44):

This is a big one where number one, if you don't have professionals in that space at your company, it's probably time to start investing in someone who deals with your data governance. And this is also where going all the way back to our beginning conversation, if you pick that business use case, picking one where the data is less sensitive to start can be really helpful. So like the call center, you might want to have it have a trigger where when something comes in and someone starts to give P, you trigger it to somewhere else. And so all data that isn't personally identifiable information or high risk data. So really you want to test those agents in places. One of the great options is with your actual real agents because a lot of the data that they deal with less concern there of what information is being passed around.

(24:33):

But a lot of it is all about are you setting up your infrastructure to mask certain variables as they come in? Are you making sure people's social security and other information is protected? And also if you have data boundaries within your company, that can be hard. This is the first time we started to hit this. Can underwriting know everything that claims knows? Can claims know everything that underwriting knows? Do you have an agreement with some agency where certain people can't see the data at your company? And what are those limitations and how do you think of them is a big question for each company to answer. And the NCCI has given some guidance to us about how we should think about it, but there's no real concrete that this is the way you have to do it. So it is a really big question of do you have people starting to have that governance conversation and what is the risk tolerance for you and your company and what's the security you're putting in place now that these tools can access the data?

(25:32):

Are you worried that they will share more information than they should? And we've seen that in a few different industries where sometimes people have asked bots to tell them private company information and the bot has responded. And so are we also putting those roadblocks to prevent the bots from doing things? I'm sure people remember there was a big, I think it was Canada Air or one of the Canadian flight companies where someone convinced the bot to give them free flights. And so people are going to try to get the most out of what they can. And so we just have to make sure we're putting those limitations and as things happen, knowing that sometimes we're going to make a mistake and how do we recover from it? How do we make it adapt? So not having that be a panic button and a stop, but more of a, okay, that didn't quite work. How are we going to reevaluate it?

Patricia L. Harman (26:18):

How heavy a lift do you think it's going to be then for insurance companies to kind of figure this out? Because you can't operate in your silos anymore because of the way that this technology works. There has to be collaboration and sharing of data and other information.

Jamie Warner (26:37):

It's definitely going to be a huge change. And we've been through a lot of changes recently. I think the thing that will also be really interesting is a lot of people are buying this and bolting it on not building it themselves. So we're not all tech companies. We don't all know how to build things. So a lot of times we're buying stuff from startups and the startups are really good at bolting on a specific area of the company or a specific system, but they're not as good at integrating across all the data sources. And so that's really I think where one of the limitations we've seen with, we see a lot of the startups InsureTechs coming in with these really grand ideas of like, this is how we would do it. And they've forgotten the fact that we're so disjointed and siloed and sometimes actually even different companies handle different aspects like different physical companies.

(27:28):

So that's a place where it'll be challenging the other places, plenty of companies don't own all of their data today and might not even realize it. So if we think about the tools in the life insurance space, a lot of life insurers use tools from the reinsurers like Munich or others for their underwriting. So do they own their data or do they have to go buy it back? Do they just get summarized data back? And so in a lot of these cases, if you do third party management for your claims, do you own that data? Do you get it back at the individual level or do you just get aggregated? Do you have to pay them for that additional service? And so those are things a lot of people, as we've built, we've bought tools and now we don't realize, oh, that data is actually inaccessible. It's in a tool with a vendor or this data is managed outside or this data is managed in this way. So that's going to be a big thing that I would start cracking the nut on, if you will, to try to discover what are the places where we have it easily accessible because that can also drive, okay, well we should start our business case here and then over there we should tell that vendor, Hey, we really need this from you. What do we need to be giving you to make it work?

Patricia L. Harman (28:33):

I had no idea. I didn't realize, I don't know what I thought, but I didn't realize that data could be spread in so many different places that insurers use. Wow. And insurance companies are notorious for having vast amounts of data and data collection and storage is certainly going to change with ai. You just kind of mentioned that. Do you see it as a strategic shift in what data is collected and how it's valued for the outcomes that it could produce then?

Jamie Warner (29:11):

Absolutely. Previously if we had text data or we had image data, it was kind of nice to have for projects where you would do innovation work. Right now that data is, I would say some of the most critical data and we have to figure out ways to store it. It's bigger, it's bulkier, it's more annoying to analyze. It's a different skillset. And so we really have to think about where do we store it, how do we store it? And let's say you're building a model, you need a lot of times historic data that's labeled. And so how do we label it? And so this is something that's interesting with for example, like the roof vendors. I was talking to one of the roof vendors and I said, how are you labeling your roof data so you can build predictions and understand it? And they said, oh, well, our data scientists are doing it, and that was an absurdly expensive way to do it. In my opinion. I would pay roofers or you could get various labels. So we really have to think differently about how we store it, and then if we want to reuse it later, we figure out a better way to analyze it. We can't just have it pushed off somewhere. We have to have it more accessible. And that also begs the question of some of our records retention policies. A lot of them are based on audit, not based on

(30:25):

Do we need more volume of losses to be able to build this? And I'm really curious to see how the data vendors are going to work with this because we've got vendors like Success for example, or TransUnion and all these different vendors that a lot of companies work with that aggregate data. I haven't really seen them start to aggregate things like some of the image inspection data, things like that. But they're going to have a really big opportunity to clean up input data for us. And I have seen some startups in that space, so it'll be interesting as it transforms, but it's definitely changed. What's the most important? It used to be the most important one with name, age, the kind of core features, and now all these other things are more predictive of risk. My age is not nearly as predictive of risk as the state of my home. And getting that information out of the pictures can tell us so much information.

Patricia L. Harman (31:22):

When you're talking about storing data, I'm thinking, well, they can store it in the cloud. But then I know someone who works at Facebook and they have incredible data farms, they collect amazing amounts of data, and we're talking that their storage facility sizes the size of multiple football fields. So how are carriers and the companies that they're working with storing their data then what does that look like?

Jamie Warner (31:54):

Yeah, this is a huge challenge and I'd say there's two pieces to it. The first is our current ideas for how we store data. I saw a meme the other day that I thought really encapsulated it. It was like a soap dispenser only. They'd stuck the dispenser into a bar of soap and they said, this is your infrastructure and this is your data. And it's like you can't just use the new tools on your old infrastructure.

(32:21):

I think the secondary piece though is really there is an actual cost to this. We like to talk about the cloud as if it's this infinite solution. Infinite capabilities, servers can be expensive, storage can be expensive, and it also involves risk. And we're really good at measuring risk, but there's a risk of where you put your data, who has access to it, and the risk of other people getting access, bad actors. We've seen Philadelphia insurance, we've seen Erie Insurance recently being taken down by cybersecurity issues and ransomware. And that's a big risk if you have a lot of personal company information and data out there, that's also data that can be acted on. And so it's a really big question that I don't think we've completely solved at this point, but I think that goes back to there's a great Jurassic Park quote, Jeff Goldblum.

(33:15):

They were so busy thinking about whether they could, they didn't think about whether they should, and we should really think about our data that way because it's not that we should store everything. It's that we should store things that we can tie to a use case that's important. And then we should keep in mind that if we don't store the right things, number one, we can start later. Number two, we can buy it. A lot of the smaller carriers are used to that where we have to go outside and buy data when we want to do a new product or have a new idea. I think the larger insurers aren't as used to that. But when we're getting the volume, and if you think of the volume, if I think about you, Patty, historically, maybe a decade ago, if I wanted to collect your heart rate, I could maybe ask you to do it once a day.

(33:59):

So max 365 data points, but probably not that many. Now I can from a watch, get it every second. So how much of that data is actually valuable? Maybe it's the outliers, maybe it's the trend information, but deciding how you store it and what's actually useful is going to be really important. Otherwise, you're just going to have tons and tons of data. And then the final piece of that is governing it. If it's labeled well, if people know how to find it, if people know how to access it. And the documentation is so, so critical here because the more you have, if it's messy going in and it can be messy in a data lake, they can pull whatever they want, wherever they want. When you try to pull it out, it's just a disaster. So really making sure that if we're going to store this, let's document it. Let's make sure we understand what the fields are and what they do and what they mean so that later when someone wants to use it, it can actually be useful.

Patricia L. Harman (34:52):

Right? It's like you still have to have a good filing system for whatever you're going to keep.

Jamie Warner (34:57):

I forget what the library code I had to learn as a kid was, but like that. Yeah, Dewey decimal system. Get it in there.

Patricia L. Harman (35:03):

So it's been said that data will become the realtime copilot for AgTech AI systems with the power to fuel realtime decisions and continuously inform the next best action. Practically speaking, how rigorous will the process be to get to this stage of AgTech AI usage? Then

Jamie Warner (35:24):

I think we're there and not perfect, right? But we are there. Next best action is here. Companies are using it. And next best action doesn't have to be perfect because a lot of times it's a recommendation to a person so they can say, Hey, thumbs down. But I think the best thing that companies can do is actually start to implement some of this stuff and say, this didn't work. This did work. Because if you're just waiting for it to be perfect, it's never going to be. And remember that the baseline is also not perfect. So human in the loop, making a decision doesn't always make the right decision. So our baseline isn't perfection that we're then trying to go forward from our baseline is kind of whatever we're doing today with our biases, our beliefs versus this recommendation and information. So I like to say, get it in the field.

(36:10):

Start testing and learning. And if it's bad, that's great. You get data, you get feedback, you can improve it. And that's why we start in areas where it could be lower risk or maybe with a pilot, but it is really important to just start trying it. And these tools are out there in a very bolt-on capacity where you don't need all your data to be perfect. You don't need everything to be ready. But that can help you understand, is there a business case here for me to invest to make it ready? And that's the place where I think it's really critical both for the people that want that investment and for the people that are trying to decide whether it's worth making that investment

Patricia L. Harman (36:46):

True. And I want to say make the investment with all of this said, where will human insight and intervention still play a role in how agen AI can be leveraged to drive decision-making and still have an impact?

Jamie Warner (37:03):

This is a place where I really hope people use this to level up. And I did see some of the recent studies, I'm guessing you probably saw them too, where people using generative AI in their research and work had less brain function than those that weren't. And so they kind of saw people doing research and combined research, they kind of stopped turning it on. And this is a place where this should be doing the more banal tasks or the intern tasks or the repetitive tasks or highlighting key things for you to then take more action or kind of take the action that needs a little bit of extra thought, a little bit of extra engagement or gives you more time with the customer, right? We're always trying to rush through all these things. What if you didn't have to spend four hours reading the conno documents?

(37:50):

Instead, you're able to spend that time kind of getting a better outcome and thinking holistically, is this process working? What else could we be doing? So I really hope it becomes an opportunity for people to turn on more and have the bandwidth to think about these things on a bigger level rather than I can just put my brain on the shelf, it's going to do its thing. And so obviously we're seeing mixed stuff in research, but I'm hoping that's where I can really change who's doing what and the kinds of tasks that we have to do where we make mistakes and have accidents and try to compare documents and really tedious stuff is where it's going to have the most impact.

Patricia L. Harman (38:29):

So we're kind of coming up to the end of our time, but before I wrap it up, I was wondering, is there anything that I haven't asked you? I feel like we've covered a lot in the last 35 minutes or so, but is there anything that I haven't asked you that you think is really important for our audience to know, especially as they're trying to get their data ready and begin to look at the process of using agentic ai, things that they should do shouldn't do, or at least pay attention to this because if you don't, it's going to be a problem later on in the process?

Jamie Warner (39:02):

Absolutely. I think the thing I would say is it's new to all of us, so don't hesitate to be the person in the room who doesn't know or doesn't understand, because there are plenty of vendors, consultants, people within your company that are repeating and haven't totally processed how this works. And so it's okay to ask those questions and it's okay to be confused, and it's actually really productive to do that because there's a lot of complexity of this stuff. It's new, it's coming fast. And so take advantage of being the least educated or the dumbest person in the room and really get the answers that you need while it's happening. And don't let people kind of just say, oh, it just works. This is still a tool. At the end of the day, it's not a perfect robot that is all knowing. It's a tool. And so your questions, your knowledge, your feedback is what's going to make it effective. And so I really just hope people lean in on that and ask the questions, kind of bring them. So that'd be the one thing I'd add.

Patricia L. Harman (40:01):

Well, thank you so much, Jamie. This has been such a great conversation and I think that you've given a lot to our audience to think about. This wraps up our virtual summit, agentic AI and insurance, the next wave of enterprise transformation. I hope we have given you all a lot to think about as you embark on your adoption journey of agentic ai. Thank you to our experts who shared their insights on ag agentic ai, Casey Kempton of Nationwide, Evan GR of Salesforce, Dave Vanek with Richmond National Insurance, Jamie Warner of Plymouth Rock Assurance, and Ika Camp SA of Kara. And thank you and a huge thank you to our audience for joining us today. And I'll encourage you to keep checking the digital insurance website for the latest information on how the insurance industry is processing and progressing in its adoption of Ag, agentic, AI, and other technologies. Thank you and enjoy the rest of your afternoon.