The Executive Playbook For Adopting AI In A Risk-Heavy World 

everyone. Welcome to our webinar on the executive playbook for adopting AI in a risk heavy world. Today’s program is being recorded, and a link to the recording will be sent out to everyone after the program. Questions will be answered at the end of our conversation today, and you can submit those questions in writing through the q and a function on the Zoom toolbar. I’m gonna turn the call over to Ryan Cantor to get that conversation started. Thank you. But we are joined by Ryan Cantor, Origami’s chief product and technology officer, and by Bob Petrie, our CEO of Origami Risk. And, Ryan, are you ready to go? I am. Thank you so much, Aubrey. Here we go. So hello, everyone. Welcome. I’m Ryan Cantor, Origami’s, chief product and technology officer, and I’m joined today by Bob Petrie, CEO and chairman of Origami Risk. And, really, we’re really excited to talk to today about kind of a lot of changing dynamics in the marketplace, but, really, risk professionals are being asked to move at really kind of record paces and and and and operationalize AI and adapt to kind of a complex world at an accelerating rate. And so today’s conversation is really gonna be all about what that means for the future of risk management platforms, workflow, ecosystems, connectivity, AI, all these all kinda come together in kind of a more adaptable operating model. So, Bob, thank you so much for being here and taking the time out of your busy day to join us. Wonderful. Yeah. Thanks, Ryan. I’m really looking forward to the conversation today. I’m thinking about the audience here. We have representatives from both our risk management clients and from the insurance entities, including pools, MGAs, and carriers. So that will sort of try to cover a lot of brow a lot of ground across across the the insurance ecosystem. But let me start off with this question, Ryan. What are the likely themes or high level examples of how AI will be adopted for commercial insurance? Yeah. I mean, listen. I think I think there’s a ton of of things that are going on around the around the world that we’re looking at, but we’re really looking at AI enabled workflow automation. Every one of our clients every every client has slightly different secret sauces. Every single client has a sleep, a different appetite, for AI automation. Everyone’s CIOs and infosec departments and security departments have a different kind of risk tolerance. And so, really, AI enabling workflow automation and really being able to embed that wherever it makes sense for individual clients, I think, is a big one. So configurable AI. The second one I would put in there would be really extracting unstructured data, whether it’s pulling data out of PDFs. Gone are the days of OCR where you have to have all these templates and things of that nature. And, really, AI is a wonderful tool of saying, I know what I’m looking for. Let’s go ahead and extract all that information in a in a in a real dynamic way. And so whether it’s photos, PDFs, unstructured data is great. The other one here would be, I think, a lot about AI analytics. Right? So, again, thinking about how does everyone be kind of become a business analyst? So without knowing how to build custom dashboards and Tableau or all these kind of business analysis tools, how can you really interact with your data using AI, asking questions, getting insights that help you kind of more dynamically manage your business? And then I would say, listen. I mean, we’re a SaaS platform. There’s plenty of out there, but embedding AI virtually everywhere we can into almost every feature of any platform you use. Right? Because it’s about reducing clicks, about reducing friction, about improving automation, about improving user experience. So I think that’s a big one. And then the other one will be, you know, we’re thinking about concierge like experiences, chat like experiences that sit across the entire platform that really allow you in a chat like way to kind of ask data to do things for you. Right? So, again again, instead of clicking and searching and filtering and and browsing around and and and managing all of those types of things, I think it allows you to kind of move faster by, you know and having a kind of a built in chat agent right there on the fly. Great. That that’s super interesting, Ryan. Thank you. Let let’s drill a little bit deeper. You started, I think, with workflows as one of the uses of AI across commercial insurance. What are some examples of AI enabled workflows, maybe down to the next level of detail? And please also discuss how it would interface with the marketplace. In other words, integration with other companies that that that our clients do business with. Sure. I mean, listen. That’s it is this is the fun part. And we’re experiencing it now, you know, unlocking that technology and making it drag and drop and making it kind of no code really unlocks the creative potential of business professionals. And when we think about workflow, whether it’s working in one system or pulling in data from third parties, maybe it’s LexisNexis or maybe it’s Verisk or Moody’s or Safelite or whatever, those are those are just data points. At the end of the day, a workflow is when something happens or if something happens, do this in this order, move data from point a to point b, transform it, add some AI to give you automation and insights. And so, again, whether you’re going to a an ecosystem or a marketplace of a of an integration, that’s just a way to advance your workflow and kind of manage it all in flight. So when we think about some examples, again, I talked to you a little bit about unstructured data. So let’s let’s think about reading PDF attachments on emails, extracting claimant information, date of loss, policy numbers, locations. Right? Automatically then going and completing coverage verifications, locating the policy, verifying the policy was enforced at the date of loss, verifying applicable coverages, linking the policy to the first notice of loss, automatically adding the policy limits. Again, putting them all at the fingertips. These are all, today, a lot of manual steps that people are doing over and over again. Taking a step further into the next step, you can kind of round robin and automatically assign a claim handler if you want. You can leverage AI if you so choose to conduct some sort of severity assessment, giving yourself a low, medium, and high, and even automatically route or assign it to people based on severity assessments. Complexity scoring. Again, different different people have different skill sets that might allow you to kinda fundamentally either kind of automate or automate in some cases or have manual review based on a complexity score. Lever leverage AI at the next step to look for fraud identify identifiers. So, you know, relevant information, determiners of any indications of fraud, assigned a score, low, medium, high, again, creating tasks or notes for a human to maybe review, but surfacing all those insights. And, again, these are all happening now in a matter of seconds on that initial kind of PDF submission. So we’re still all pulling all this data out of a PDF, extracting it, putting it where it goes in your system of record, and then running all of these steps against it to just surface all these insights, leverage AI to then create some initial reserve recommendations. Again, human in the loop. You can edit it. You can tweak it. But instead of starting with a blank sheet of paper, you have some place to go. You could automatically generate draft email notifications. If your company is really into automation, you can automatically send those email notifications. So, again, deciding individually where is the right place to ask human in the loop. You could generate tasks again for people to do. I mean, that’s just one example. I mean, you could escalation rules, policy administration, read PDF attachments, AI supported data validation, underwriter reviews. I mean, the list goes on and on, Bob. Right? I mean, I’m just giving you an example of you’re taking all these capabilities and you start stringing them together in an automated way, and it really is a very creative kind of endless process. And what the great part about it is, you can just start at the beginning of your workflow. First, you try, then you trust, hand it off to a human, and then, you know, a month later, you can add another step, and then you can kind of add another step. So you don’t even have to do big bang. You can kinda ease into it and build confidence as you go. Great. Any any examples that that you could show in the real world of that that would sort of illustrate those ideas a little more to to to the folks on the line? Sure. So let me let me show you just a a brief example of of an AI workflow kind of visually. Right? So, hopefully, you guys can see my screen. And if I hit play, you’re just seeing again, I talked to you a little about no code, drag drop. Right? So tiles, steps, triggers. When this happens, go get a claim that’s thirty days old. For each one of them, create a note. This is a very basic one. This is an auto glass claim. If then, in this case, you’re seeing a a a SafeLite integration there going out, doing something as part of that particular workflow. Here, you see an AI tile where we’re using AI to automatically draft an email. Here, we’re looping through. We’re creating policy questionnaires. We’re sending an email. Right? So, again, the world’s kind of your oyster, Bob. Right? These are just real life, whether they’re small, they’re medium, they’re large. You can break them down to small parts. You can string them together if you want to. But, again, visually, every business owner, every risk manager anywhere can look at this without being an IT professional and put your business head on and know exactly step by step what what the system’s doing for you. And so how does AI factor into this workflow? So, obviously, the the the to your point, I mean, AI can be included in any step of that workflow where people feel appropriate. Right? So, again, everyone has a slightly different risk tolerance for where and how to use AI. So instead of kind of one size fits all, you can put it in where it makes sense for your business, and you can kind of set it in in in in your layers. And, again, as with all things AI, test it. Right? Play with it. Play with those workflow. Run a couple of real world scenarios to it. Make sure you’re comfortable with those outcomes in a safe space. And then, obviously, you know, with one button, you can promote that and and get value pretty quickly. And and how how do do customers set their own individual risk tolerances sort of in in in in AI workflow? How how would that work in in in in in practice? Because as I think as you started off the conversation, there’s there’s a wide variety of of of approaches that our customers are taking to make sure that they’re getting both business advantages and they’re getting it safely. So we we I think anybody who’s building AI into their into their software tools needs to be very thoughtful about how how exactly their customers wanna use it and how they wanna limit their exposure and risk. Sure. I mean, obviously, if you, you know, if you read the headlines and news, everything’s agentic now. Everything’s this, and and and that’s that’s that’s that’s great. And and there’s the capability there to have straight through agents doing things for you? Sure. Is that where I think most companies start, Bob? No. And so I think, again, choosing that you have the human in the loop at certain steps. So in some of those examples, the AI may create and surface those insights and create tasks for the human to review. It’s a lot of automation that gets you there. It surfaces all those insights. But as a company, you’re saying there’s the human in the loop and doing this step. And, yeah, I think you start there. Right? And if over time, you realize and you gain confidence and you gain feedback that that AI recommendation was right ninety five, ninety six percent of the time, then you have a different decision. Then with a lot more confidence, you’ve already kinda first retried them, we trust, crawl, walk, run. You’re then able to say, Let’s just remove that human loop step, and now we do have some AI automation that we believe and we trust in. But I think that’s likely the way people are thinking about it, Bob, is that you start with a place that says, I really want it to be a helper. I really want it to be an aid. I really want it to automate what people are doing manually, but I still want a human to to look at it, review it, gain trust in it. And over time, you have the capability then of automating it. So I think that’s the spectrum and just being able to be agile enough in whatever system you’re using so that you have that real world command and control at your fingertips. And at at each client’s pace, business need, and comfort, they can turn those pieces on and off as they need to. That’s terrific. Alright. So I think at the beginning, you also mentioned using AI to do to to extract unstructured data. And in the insurance industry, oh my goodness. We’re sort of awash in spreadsheets and PDFs. We take data out of systems and turn it into a report, and then need to get it back into a system another system. And what that really does in my experience across this industry is it limits the information that can easily be accessed to people because there’s a barrier. There’s a barrier of of of typing a rekeying or the chance of of inaccurate data getting in because the tools to extract that data today are are not particularly are not particularly adept. So maybe an example of some discussion of of of of of of how AI can extract data from a from a PDF would would would be helpful to folks on the call. Yeah. So let me let me tell you a little bit about it, and then I’ll I’ll show you a quick version. And it’s a little bit of long video, so I’ll I’ll I’ll dance around to get to the meat and potatoes. I don’t wanna waste anyone’s time. But I think when you think about it, AI is really good at finding needles in haystacks. It’s the best way I could think about it. It’s the most accurate human instead of reading a hundred pages. And so the way we think about it is, at the end the day, you do know what data you want. Right? So if you were a human and you think through that what that manual step would be, I’m gonna open up a certain, you know, set of documents, and I’m gonna read them, scope them, look them, search. I am looking for something. There is a desired outcome. There’s a goal that we’re trying to achieve. And so all all you really do is you tell AI, hey. Listen. And you have the right models and the right tech and all that, but this is the structured data I want. I want policy ID, and I want policy name, and I want coverage limits, and I want brokers or carriers. And I wanna know all of this data in a in this format. And then you throw a PDF at it, or you throw a picture at it, or you throw a Word document at And the best part is the AI doesn’t care what the format is. It’s gonna go look at that document and extract those pieces of information for you so you can move to the next step of your workflow. And that really is how it unlocks this this, you know, kind of unstructured way of of of working. So let me kind of very quickly kinda show you this this example. So here we have a confirmation of insurance, real life example. If I kinda move my mouse out of the way so we don’t, see it there. We’re just gonna show you kind of the fun part of this demo where we show you, look, Bob. It’s, like, best demo ever. It doesn’t exist in the system. This data is not there. We look for that ID number. It’s not there. In this case, one of our one of our colleagues is using email. Right? So there are lot of ways people can unstructure data into their systems of record. Emails are very, very popular ones. So we’re just gonna email in a PDF and basically use AI to extract all sorts of information. Didn’t care how many pages it was. It didn’t know what that format was. And immediately now when we search, there it is. Right? And being able to pull that data out of that PDF. So now you see attachment of that PDF in real time. And now we’re pulling that data out, and we’re putting it exactly in the system where it needs to go. So now it is structured, searchable, reportable, and it is now something that AI can interact with. Right? People a lot of people wanna use AI, and maybe you’ve heard of bad data, bad AI. Right? So still having a system of record and getting that data in the system now allows you to use AI on top of this to gain all kinds of insights and automations and pieces of that that that nature, but getting it still in a system of record. So now you’re kinda double dipping, Bob. You’re using AI to extract the data and get it in the system, and then, you know, we can also use AI now to interact with it and gain insights about what’s next. Wonderful. Alright. Touching back again to the start of this call, you you you had mentioned the use of of of AI for analytics. And, you know, this is something that I’m particularly excited about because reporting in the insurance industry and really across any apps, this isn’t just an insurance. It’s it’s really a traditional SaaS approach to reporting. It’s either sort of preset dashboards or preset reports. Right? We sort of figure out what you want and need, and you can run it anytime you want with with the latest data. But now any of us that have been that have been using any of the chat tools know that there’s another way that you can that you can get answers from your data. You can just ask it a question, and AI figures out the way to do it. It doesn’t it doesn’t rely on a whole bunch of preset instructions in order to to get you there. It it it finds the the right way to do it. And if it misses the mark by a little bit, you just give it another prompt, and and it gets you there. So do do you have an example you could share of how that works in the insurance industry, maybe something on origami? Yeah. So, obviously, a little over a year ago, we launched our our TCORE module, and our TCORE module allows you to natural language and ask all kinds of really intelligent questions. So, again, getting back to the last point I made, you still have to have your data in the system. You still have to have structured data. You still have to have a system of record because the AI needs to understand what that data means. Right? And so having a a an AI analytics layer on top of it that allows you to simply ask it questions. And as you point out, dashboards are great. The way I like to say it is, but most dashboards leave you with more questions than answers. Why did this go up? Why did this go down? What’s going on? And, usually, the next action is let’s go create yet another report. Maybe the report is gets emailed or it’s an Excel report. And what do I do immediately? All kinds of manual analysis on that data and slicing it and dicing it because I’m trying to uncover the root cause. And so the idea here is just to accelerate and kind of cut through all of that and just ask it the question that you’re trying to understand. Why did this go down? Why did this go up? And AI understands the data that’s underneath it, under looks at the content of that data, a SQL query, which is a little geeky, but it does, and you can kinda try try it and trust it. It goes and queries that data. It pulls that data back, and then AI does another pass at it, Bob, and says, here’s what the data looks like. Here’s how many rows. It’s date limited. Here’s the kind of makeup of this particular answer, and it chooses the best visualization to visualize it on the screen in a way that makes meaningful sense. So I don’t have a video of that right now, but I think the the concept, though, is is pretty is pretty is pretty powerful because it allows you to interact with your data as a human. Right? It allows you to have a human to human conversation and just ask a question. So if you’re a risk manager coming out of an executive meeting or sitting in an executive meeting and you’re getting asked a, you know, a question he hadn’t thought of or what’s next, today, answer might be, no. Let me take that back, and I’ll get back to you in a couple days, and I’ll run that analysis. You could do the question live in the meeting and have an answer put it up on a put it up on a on a Zoom presentation like this in five seconds and say, here’s here here’s what our data is telling us. And I think getting faster insights, being the person who’s in the know, it it it sends all kinds of wonderful signals that I’m in command of my shop. I’m in command of my data, and and I understand what these things mean. So I think, again, unlocking those tools is still predicated on having data instruction systems. So, again, you can you can take advantage of these things, but it’s it’s truly remarkable what what what we’re seeing right now. Super cool. Maybe the last topic before we get to the end and and and get to q and a. You you mentioned at the start of the call a a concierge, and I’ve seen versions of this in applications that aren’t particularly focused on insurance. But was sort of curious about how you think a concierge might help across a claims or a policy administration or a Remus or or a safety system, you know, to to to add value to users, make the system easier to use, allow you to do complicated things in a simpler way, any of those any of those themes that that that resonate with you? Yeah. So if you think about the journey we talked about with workflow, mean, think about those are structured workflows. Right? When I press this button or I sent this email, a company wants command and control of all the all the things that are gonna happen kind of in an organized way, and that’s great and highly configurable and benefits for those particular use cases. But then there’s the day to day, Bob. There’s the user in the system, and how do we make them more efficient? Right? The clicks, the browsings, where to go to find things, where to get insights. And so if you think about a chat experience that almost at a user level allows just them to be kind of, you know, kind of a lightning machine, right, where they can become more efficient because they have a chat interface built natively into the system. Not bolted on, not outside the system, not consuming all these kind of extra tokens and all this kind of stuff with weird connections, but native to the system so it understands the context of all your data. It understands the context of how you use the platform, and it interacts with you in a real way. So let me give you a kind of a a a quick example of something that we’re we’re gonna be launching here. Let me see here. Right? Make it make sure I picked the right one. And we’ll launch you this in August, I think, our next our next major release. Yeah. So let me yeah. So this is let me go to this. So if I make this a little bigger so everyone can kinda see it, we’ll go back here. And so if I if I make this a little bigger so people can kinda see it, this is our our kind of a new concierge application. We’re gonna do some basic ones here for a minute. You know, take me to that particular claim. You can go click on the claim, see if you’re button. So now we’re self navigating. We’re asking questions. Do I have a claim in the system? When you need a little bit more space to work, you can go ahead and do that. It, you know, becomes nice and nice and big if you’d like to do that. Have we already sent a letter confirming the receipt of the claim? It’s gonna go ahead and tell you here all the attachments. Why don’t we just throw a microphone in there, Bob? So if you don’t even wanna type and make it even easier, you can just talk into the system, ask it questions, it’ll kinda respond to you in real time. And so in this case, you know, can you craft a letter for me? In this case, you know, in this case, we can kinda go ahead and create a task. We’re gonna let them know that we’re in the central time zone. At the moment, it’s gonna go ahead and update that, allow you to create the task. In a second here, we’ll show you exactly what we did. We’ll refresh the page here on the right hand side. There’s your task that’s created for you in the system. Skills. So as an administrator, you may wanna unlock capabilities for your team, different pieces of things here. We’re showing you kind of comparing policies side by side. If that’s a skill that you wanna do, we’ll have I think we have thirty, forty, fifty skills out of the box all ready to go, covering all sorts of the platform. But, again, at your fingertips, you don’t have to go search for a policy, then search for another policy, then export them, then do an analysis. This is just a chat like experience where you get to interact with it in kind of a real time way. We’re recommending skills based on the page you’re on. So, again, these are all kind of customizable con control settings. But, again, you get the you get the gist that this is a really powerful chat experience that allows you to navigate, find answers, build tasks, automate workflows, and really anything a human can kinda do in Origami. You now will have this kinda built in chat assistant across the entire platform that will help you do it for you. Terrific. Alright. We have just a minute to go before q a, but, Ryan, maybe you’d leave us if there was one sort of thought that you wanted this audience to to to come away from this this conversation with, what one one major takeaway. What would that be? Listen. I think AI is obviously everywhere, and it’s innovating, and it can be scary, and there’s a lot of weird headlines. But here’s the best thing I can it’s if you you’ll be impressed if you allow yourself to be impressed. And so that also means setting up a space to try things and experiment. Bob and I have this conversation all the time. Well, he says, well, could AI do that? My hands I don’t know, but I don’t see why it couldn’t. If a human could do it, let’s go figure it out. And that’s kinda the frontier that we’re at. And so I would say, you know, have a safe doing environment. Be willing to try things, experiment. Well, what if we did this? What if we put AI in the workflow? Okay. I didn’t really like that. Move it around, tweak it. Right? That’s where kind of real cool things happen. And and so I would say, you know, whether it’s a marketplace of kind of inspiration, things that other people are doing, webinars like this, white papers, be inspired. Think think about how people are using things, and then I would just say in a safe space, try it. Be willing to try it. Be willing to put yourself out there. You’ll be amazed at what it can do, and it will unlock a lot of things that everyone may kinda come into this. Say, well, I don’t know if AI could do it as good as I do it or AI could do it as good as so and so could do it. Just try. You’d be you’d be really surprised. And, it’s not gonna be about replacing you. It’s gonna be about making you a superstar. It’s about making you efficient. If you could ten x yourself, that would be a really smart decision, and I think that’s the right mindset to put yourself in as you’re thinking about unlocking AI capabilities. And I think that’s exactly right. Alright. Aubrey, would you, lead us, through the q and a if there’s any questions from the audience? Yeah. Of course. So we’re opening that up now. Audience, if you still have questions, you can type those into the q and a section. But we have quite a few to start. So I’m gonna start with our first one. How can you address end user change management and adoption resistance for these tools? Well, why don’t you take that one? How do you so first of all, I think change management, and I do think, again, I think all the tests and the experiments you do in a lower environment, I mean, record them. I think all of those tests help you with change management because you get to show real life examples. We put this many PDFs through it. Here are the of it. Here’s the good, the bad. Right? And and you view yourself, build confidence as a leader. So I think that’s part one. Part two is I do think there’s really two ways to think about it, and and I will see the most success I’ve seen is you know, and this might run a little bit contrary to some culture cultural norms, but a top down AI strategy has proven to be more successful. And what I mean by that is as a leader, you’re getting comfortable with it first. You using it, being hands on, understanding what it can do gives you a sense of confidence and and assertion that as you roll it out in your organizations that that that you know something to be true. Right? Allowing the individual stakeholders who might have varying degrees of it’s, you know, it’s it’s new and it’s awkward, so I wanna tell you why it’s not as good as what I do. There’s a lot of defensivism that can come to that. Has to start with you getting comfortable that they can do this, and here’s what it means for your organization. So I would just say from a change management perspective, start small, crawl, walk, run, build confidence, and set a culture where you encourage innovation so people understand that they don’t have to be scared of it. They can use something like concierge that I showed you and build a skill or show you, wow. I was able to do this really, really cool thing. And it’s a way for them to be part of that innovation, not that the innovation’s happening to them. Great. Thank you. Okay. We’ll go to our next question. So Caitlin asks, is there any way to use AI to document emails directly into the notes section when using Origami for claims? Say that one more time. I So is there any way to use AI to document emails directly into the notes section that they’re using when using Origami for claims? The answer is yeah. That would that would absolutely be a workflow. And you think about, you know, emails either that you’re emailing into the system. We also have a new Outlook plug in that that we just launched in May as well. And so I think between those two pieces, the email becomes a body, becomes a system of record. And so if you wanted to create a workflow that as an email gets added to the system, I wanna extract again, maybe not helpful to extract the actual email, but you wanna extract certain pieces that might be in that email and put it in a kind of a predictable standard structure, task or a note every time. That that’s absolutely doable with a workflow. You’re basically taking specific pieces of information using AI to identify those pieces of information and then writing or creating a task or a note with those specific instructions in that order. Okay. Thank you. And I think that relates to, another question we got, which asked about they often receive claim forms as PDFs attached to emails from their members. So could the AI then extract and use that data to open a claim? It is that is exactly what it does. I mean, that is a picture perfect example of the the the test I did with Bob, which he he’ll laugh at is I gave him a piece of paper, which he wrote sideways on just to be cute and funny, and we took a picture of it. So we were able to extract Bob’s terrible handwriting from a picture sideways and and put that as structured data in a system. So, yes, to your point, struct PDFs, unstructured data, the extract information, give me your tired, your poor, give me your photos, give me your written on napkins, give me your PDFs, extracting that data, and put instruction in the system. Absolutely. Okay. Thanks. Now it looks like we just have time for oh, sorry. Go ahead. Yep. Yeah. Let’s do one more, Aubrey. Okay. So our last question is gonna be, where is that policy data housed after the AI tool processes the data? In or in Origami is the short answer. We’ve got policy tables. So if if the alternative is you type it in yourself, you you open a policy, you open a binder, you look at the information, you type it into the right field. But with with the extraction, it will take the information and put in the right place in the right tables. And with that, I think that’s an awfully good place to end. I guess I would like to say on behalf of of Aubrey and Ryan and I, thank you so much for for for joining us. We had a lot of fun thinking about this. I’m not sure if there’s a poll at the end, but if there is, please tell us what you thought of this. We would we would love your feedback and certainly appreciate you spending half an hour with us. Thanks so much. Thanks. And as a reminder, we’ll send you all the link once it’s ready. So thank you again. Have a wonderful day. Thank you.