From RPA to AI Workers: Appian’s AI Playbook

0:00AI is not replacing underwriters or claims examiners and adjusters. Yet anyone that you may hear that tells you that it is, either they're naive about insurance or they're trying to sell you something, but AI is fundamentally changing what those roles focus on, and it's increasing their effectiveness at unprecedented levels. Welcome to the Talking AI Podcast, where we talk AI with both experts in the field and early adopters. I'm your host, Matt Page, and we're here to demystify AI for you so you can get some value from it. Let's talk some AI. insurance is one of those industries everyone knows could use a little bit of change but few are willing to actually disrupt it and today's guest is doing exactly that so we have jake sloan vp of global insurance at appian is a leading the change in modernizing things like underwriting and claims through agentic ai and building systems that don't just automate work but actually think alongside people and we're going to unpack how he's approaching ai and one of the most complex risk sensitive industries, what pitfalls he's seeing firsthand from hallucinations to data obsession and why real transformation starts with purpose-built domain specific AI, not just a bunch of hype, but welcome to the show, Jake.

1:15Yeah. Thanks for having me. It's great to be here. Yeah, this is going to be a good one. I love this whole idea of very purpose-built domain specific things because everybody sees the flashy open AI just launched their new general purpose agent, but there's a lot more to be had in the domain specific stuff. There is. I like the interest is all around us, whether it be your person, your vehicle, your property. It's something we all live every day. It really is. That's, that is the truth. And Appian has been around for a while. So since 1999, I was actually doing research before the pod. Yeah. It's, I mean, it's crazy to think about far away.

1:491999 actually is now, but it's a whole nother topic. But you've been doing Appian in general has been doing this process automation before it was cool. Right. As the process company, what are the learnings, the parallels do you see from this traditional process automation, RPA versus what we're seeing now with this agentic AI that's coming in on the scene? What's different about it? The same? No, it's a great progression, if you will, and the evolution of it, right? Our four founders of Appian are still very much hands-on. And I hear stories, of course, wasn't there when they founded it. yet I've been working with them now eight years as an employee, three years prior to that as a customer.

2:31And we see how the industry has evolved from KBPM, case management, and then RPA and the promises as far as what we did deliver, maybe that fell short from what RPA had expected from the business case. What I'm excited about most and where we are at very, very quickly is the most significant, I believe, inflection point within the industry since digitization began. And so when you think about how interesting it is, how most insurers are, they aren't failing because of the tech. They're really struggling because they're trying to apply, say, general AI solutions that you hit it with your open. Like the latest GPT model is out.

3:07It's a great search engine. It does really cool things, but it falls short as it relates to specialized, highly specialized processes for, say, and running things for insurers. Yeah, totally. And I think the other side of it, too, that I always go back to is obviously AI, generative AI. It's probabilistic in nature, right? But if you think of traditional RPA, you're having to specify every single thing that's happening in this process for it to execute correctly. And now you've been given this like magic gift. I'm assuming it feels like a magic gift that maybe it can bite you as well. But that can actually make decisions based on context.

3:46Yes. What does that look like in terms of how you think about integrating this new capability into the existing world of RPA? Yeah, yeah. What is that like? You took me back to memory lane there, right? So I remember as an operations leader for a large insurer, we had RPA and the Accelerator. We started with SQL databases and access. And if you look at RPA, when your desktop would log out, now imagine you have, and the bot would stop and nobody had eyes on it. We had bot monitors. We had 10 bot monitors for every two bots. And so the efficiency curve was upside down. And just imagine this world now where you're at, as you noted, it has such power, such content.

4:27The context is probably the word I use the most when we're talking about all these capabilities is that, just like you said there, that, you know, something that can contextualize and understand what's being ingested, the context of how it's being applied. And so when we talk about AI, for instance, in insurance, we've got to be very specific about how we quantify both what the context is, the value that the context delivers us. And that's a massive difference when asking chat GPT to draft an email versus deploying AI that can autonomously, contextually review, say, a commercial property submission and other facets.

5:04Yeah, totally. And the one piece too, you mentioned earlier, like you get to this, while it's both powerful, but it can hallucinate. Everybody at this point understands the concept of hallucination, I believe. But when you get into these riskier domains, like the stakes get higher, right? And the problem I think a lot of times is AI is great at giving a very plausible answer that sounds very good. I equate it to my nine-year-old daughter. She's getting very good at saying things that seem right. Now I'm like, are you telling me the truth or what? I don't know. Quick break in the pod. If you're listening to this podcast, chances are you've been thinking about how to actually use AI inside your business.

5:47And that's exactly why we built the AI Opportunity Finder. It's a free tool that helps you uncover high impact, tailored AI use cases based on your business, your goals, your pain points, and your industry. No fluff, no generic use cases, just real ideas that fit your business and the rank by ROI potential. It takes about three minutes to run and it's like having your own personal AI strategist for free. If you want to try it for free, check out the link in the show notes or go to hatchworks.com backslash AI-opportunity-finder. How do you think about that? Because obviously guardrails come into play and things like that.

6:23Because if you get something wrong in the insurance underwriting space, that's not good. No, absolutely. Yeah. And I have four children. I have an eight-year-old daughter. And the conviction at which they can tell it, it does, it makes you question it. And just like the hallucination, when you start talking about that, even the hallucination, I was getting so convincing and such conviction. I've even seen fake URLs that look absolutely believable. Yeah. And so when you start thinking about how you approach this, You've got to be able to, and maybe a place to start is the clarification between agentic AI and being very purpose-built for say, underwriting and claims or enterprise document ingestion or enterprise adoption, being able to give it a specific set of goals that are very intently defined around say, underwriting and claims in contrast to that just very general, here's an answer to add that you see.

7:13And then of course is an analogy for what you can just the general GPTs and LLMs. Yeah. And I think that's the key thing. like you're hitting on earlier and I'm guilty of this too. When I see open AI, man, that's clod perplexity. They're coming out with these cool general purpose agents that can like order Uber eats for you. Yeah. But when you, but it's not purpose built for what you're doing. And I think there's two angles to this. A general purpose is cool. It can do a lot of things, but if you want it to do something specific, your error rate, your blast radius, whatever you want to call it is wide.

7:48Like with us, when we talk about agentic AI automating things, the more confined you can make the process, the workflow, the thing you're doing, the better, right? Because then you have more control over this probabilistic system and you can add in additional guardrails and things like that. But I think that's the key. And I'd be curious, like y 'all are starting to build some agents, agentic solutions around underwriting claims, things like how have y 'all approached that being built in this kind of purpose built fashion versus let me go grab open AI chat GBD off the shelf. Yeah, no. So we started with maybe reverse engineering, maybe too fancy.

8:30But if you think about the knowledge worker today, because it's something we should level set maybe from the beginning, we see that there's going to always for now for the foreseeable feature, B, an expert human underwriter, claims examiner, claims adjuster, they're at the console that's made better by this. So we approach it from a skills assessment and we look at this and there's 25 skills of an underwriter. We look and say, what do we believe is best served by entry level work? What's maybe a mid kind of complex work and what's highly skilled, highly specialized work. And so we've taken that approach and we've unpacked it to then go in.

9:08And so if we're talking about this maybe in the real world about a commercial underwriting application, traditionally that underwriter might spend 60 to 70 % of their time gathering data, validation, pulling information from submissions, checking against lost runs, looking at validating property details. Let's just for a second imagine the old world that didn't have this capability. We didn't need highly specialized colleagues and staff to do that. So we came up with things like OCR, maybe we had different data centers that are just input only. And that's how we've approached it, Matt, in the context of those lower, and now we're moving from lower to mid-skill kind of skill sets to then automatically extract data from, say, unstructured submissions, process reference that against maybe multiple data sources simultaneously.

9:54And that's how we've approached to bring the value. Because cool is there, the fascination is there, but it all means nothing if we don't quantify the value from the start of what it brings to organization. Just like our, maybe our offshore day entry strategies or our offshore rendering strategies, we've got to have that value from day one. And that's also something we keep top of mind at Appian. Yeah. A really interesting point there, starting with the skills and breaking that down. I hadn't thought about it in that fashion, Cause then you can make it more composable from there. I don't know about you, but I feel half the time executives, prospects we're talking with, they come to us and they say, I want, I want an AI agent.

10:41And you're like, great, that's awesome. And then the next question is, what do you want the agent to do? I don't know. And so what is, cause I'm sure you get this too, and you're obviously in more of the C-suite, but there's a lot of the top down pressure from the boardroom, from executives saying, we need AI and the people doing the work are like, great, but what, what do you want to do? Talk, talk about that kind of working backwards. You hit on a little bit, starting with the skills, but how do you go to actually the workflows and the processes and things like that and identify, okay, what's a good case for automating something with AI versus not?

11:21Yeah. Great, great question. You mentioned. Yeah. A couple of thoughts on it. First of all, at Appian, we believe that AI belongs best inside of a process. And we've been referring to them, especially the agentic capability, as an AI worker. When you put AI to work with the AI worker in a process, say underwriting your connected underwriting solution or connected claims or claims solution, that agentic underwriting agent or the agentic claims agent goes to work for you, you quantify the value. The second aspect of this, you nailed it. We have seen the board, as they should, put a tremendous amount of pressure on the C-suite to really be able to bring AI.

11:58They see it and they're personalized. They see the search engine. They're blown away. But then they see how these other companies, and there's some fantastic companies and fantastic results out there of how they're achieving it. And I don't want to pick on an A industry, but I just highlight it. The life industry, for instance, has been slower to adopt. And that's okay. So more conservative, but we literally have examples where we will have organizations that we've gone into six, eight weeks ago that said, do not bring AI in. We do not want to discuss it. If you have your workbench, we want you to be able to give us a toggle to shut everything AI off.

12:30What? Nothing to do with that stuff. And then three or four weeks later, they've called us back and said, hey, the board can't get AI fast enough. and it's just been really fascinating because if you start to look at those organizations and we all know tech debt and this is built on decades legacy old systems if you think about if you're not doing something today just imagine how fast things are moving what a year looks like much less two or three when you get around to really getting your comfort and setting a pace just how much further you're going to be behind and i don't say that as a shock i say that as a practical c suite leader, if you aren't embracing and comfortable and understanding.

13:06And then, you know, and at some point, like if time prevents here, it would be great to talk about the culture because the culture has to also be flexible and adaptive if you've got older culture, you've got to really be focused on that too, but long, long answer, but that's how we're approaching it. And let's get into it because the blockbuster versus Netflix analogy has been used a ton, but like. That was very vertical in terms of that disruption. AI is everything. It's horizontal. It's horizontal. It's every domain function. It's like diagonal. It's going three-dimensional in a sense. But you mentioned earlier that apprehension, you mentioned life.

13:42I don't know if those life insurances or life science specifically. Life insurers, yeah. Life insurers, yeah. So what do you see as the biggest apprehension? Have you been able to dig into why people are apprehensive there? What are the key things or reasons? Yeah. Yeah, there's kind of several consistencies. One is just understanding what it is and what it isn't. And we've seen some horror stories out there. We've had maybe that maybe they've had a personal, a board member or a C-level leader that have had maybe an interaction. You know, and I have one of the time permits I can share. I have a car that has a crock inside of it.

14:17And industry leader, I started asking questions about life insurance and the hallucination was epic. So maybe they've had something like that to inform their view. But I think it starts with that. it also starts with the fear. Your staff, your organization, as you should be loyal to your people and your organization, there's this fear maybe of, hey, maybe I don't need a third of my organization, as well as then the technical complexity that goes with it. We have to do it. We advocate for what we call private AI, which is AI that stays inside of your organization. It benefits from the outside models, yet an Appian's capacity stays inside of our platform as a service.

14:51So all that data stays resident in their four walls. Even within certain regions, You can define that through our partnerships with AWS for hosting. So I think in summary, it's maybe just the lack of understanding as well as the maybe change is hard. And so risk averse or change averse organizations, it's tough to come in and just say, yeah, let's just put rocket fuel on change and go from being maybe a waterfall organization that was very comfortable with a slow pace. Maybe they've made millions, frankly, and you've got good risk postures, good profitability. Why change it if ain't broken kind of thing in contrast to the modern insurer must adopt these new capabilities to stay relevant, to stay profitable, to be around for the next generation.

15:37Yeah. And you mentioned the change management side of it. I think so many companies and executives and IT teams, they kind of say, okay, we've given you enterprise co-pilot, ChadGBT, whatever. All right, we're good. And then month two later, why is nobody using this thing? Because it's change management more than anything, because you have to learn a new way of working with these tools and capabilities. Because most people stop at about making it or using it to help improve your email or something like that. But that's like the tip of the iceberg. Any thoughts? And this is to that culture point.

16:14How do you bring the organization along with you? The obvious thing is, oh, let's show them use cases. but sometimes that is you get the fizzle or it's the excitement and then they go back to the wrong way. Any thoughts around that aspect of it? It's the human element of it, right? And so when we think about - Because humans don't like change, right? We're risk averse in general. That's in our nature. Exactly. You're about right, tech. I love change, but then at the same time, I'm like, I don't really know if I want you to change this. And it does, it forces you to change. And I think when we start, especially with the replacement fear, right?

16:47When we think about how we, from the Appian leadership side and our industry domain team, we say that AI is not replacing underwriters or claims examiners and adjusters, yet anyone that, you know, you may hear that tells you that it is, either they're naive about insurance or they're trying to sell you something, but AI is fundamentally changing what those roles focus on and it's increasing their effectiveness at unprecedented levels. And so if you put together an AI adopter, or if you put together a small group. And when I was in one of my senior technology roles, I like to pick individuals in the organization that thrived on change as well as those that absolutely avoided it.

17:29And that way you've got a good diverse mix. And we've seen time and again, that bringing these colleagues along, letting them have a say in understanding it. They know the market, they know the market deeply. They know the relationship, all the different things that make them successful organization. Now, how can you bring that AI to work in a process with them as a focus group? And then we do, we see organizations that will compare and contrast those that are using the application versus those that are not. And again, I'm being fully transparent to a fault here. We have seen time and again, that approach allows them to really rise up and show and leverage that AI and process to value, get rid of that administratively burdensome work, give it guardrails, and it's not perfect every time.

18:10And that's the point because you can continue to tune it, but those colleagues that are helping influence that and almost an AI steering committee kind of capacity are winning far greater than those that are forcing it top down, expecting that you and I are just going to soak up change like that when in fact we don't. Yeah, totally. I always go back to the Shopify, the memo that got leaked and then the Shopify CEO said, I'm just going to put it out there. And the key thing that's just stuck with my, in my brain, since I've read it is this term called reflexive AI. And it's basically where it becomes reflexive, just like a reflex.

18:43When you go to do something, you're not thinking, oh, do I use AI? Just like you wouldn't say, oh, I'm going to use electricity today. It just is second nature. That's right. And the other point was he mentioned the best way to learn AI is by using AI a lot. Yes. And I think a lot of people have this artificial barrier of I'm not technical or I got to do all this things to, before I start, no, just start playing with it is the best way to start incrementally building over time. Exactly. I mean, you take that and who best to do that than those colleagues that are there. And we see time and again, the nobility of the industry, whether it be a domain underwriter, whether it be a claims examiner, don't forget at the end of the day.

19:24Yes, we talk about it in terms of profit and loss and everything, but don't forget about this is a human life that we've insured in a family that we've helped provide for, or this is a large commercial undertaking, a property that we've provided our business for disruption or workers' compensation. So if you come from that place, which everybody eventually does, and you figure out what parts of the role are better served with this incredible capability we have in front of us, then let them drive. Let them drive. You bring your architects, you bring everybody, put some safety switches in there, put some different temperature controls to dial it up and dial it down.

19:58And that's something that's working exceptionally well. Yeah, you mentioned the fear of the job loss piece too, But what I always go back to is like the to-do list of things that you are never able to get to is so long. There's always more things to do. And this is only going to enable you to do more of those things. And back to the Shopify memo, he mentioned too that a lot of his people were accomplishing with AI things that he never thought possible. They go, they're pointing AI after these big, hairy, audacious things that not only did they not have time for, but they were big and large, which I think is just an interesting point as well.

20:38How can I work with AI versus it being like this competing force? That's right. And if you look at that BHAG approach to say, say the organization, I want you to straight through process without sacrificing profitability, relationship. We can't change any of that, but I want you to increase straight through process by 50%, 80%. I had a claims leader that we talked to this week for one of Australia's largest insurers that said, I want 100 % straight through process, all claims. So now, is that a reality? No. And this person knows this yet aspirationally, if we approach the business with that innovative mindset every day about how we're going to leverage AI, how we're going to use it inside of a process, the human AI partnership model, where we can take all that, that data intensive, repetitive tasks, shove all of that into the generative agents.

21:29then we just get that much better every day. And this organization has such a phenomenal execution culture. And this leader in particular is just doing incredible things with our organization because they realize that this is how you can approach it. That sets the tone. And then if you get to 40%, 60%, 70%, that's strides that are just massive. I think it's interesting too that a lot of people set such a high bar and expectation for AI. It cannot be wrong, right? But if you're comparing it to a human doing the same thing, what is that failure rate? So I think a lot of times folks need to almost take a logical step back and say, okay, in whatever process it may be, maybe we have 85 % success rate with a human or 95%, there's some 5 % that's wrong.

22:19That should almost be the baseline that you're working towards versus perfect. Cause if you're trying to get perfection, I know you mentioned it earlier, but on the flip side, if you try to get perfection, you may never start. I think it's like perfection as the forcing function is good, but if perfection prevents you from actually leaving the starting line, then it's like counterproductive as well. Yes. This comes up often where we'll hear a lot of, oh yeah, we have our strategy. We're working to cleanse our data and we're going to get that started in insert timeline here. It's never less than six months.

22:53Most of them are like behind the scenes, hey, it's 12 to 18 months. And that's just, you know, fundamentally flawed in the sense that from our perspective, we're full agile and iterative where progress is far better than perfection, right? And one of our founding company values is to really focus on progress and not perfection. Because the second that you move that forward, even 1%, you're 1 % better than you were yesterday. And one of the biggest mistakes that we see is that we're going to spend 18 months building the perfect data strategy, then we'll implement AI. Meanwhile, their competitors are behind the scenes, they're extracting value from imperfect data using guardrails to maintain integrity.

23:33And it's that whole learn. And the other thing that's incredible about this, by the way, and your audience gets it likely, when you or I as an underwriter, let's say, you're expert, maybe I'm junior up and coming, we both benefit the model by when our knowledge contributes to that. In contrast to today's environment where maybe I have to be shoulder to shoulder with, maybe I have to go and do training enrichment versus the model. Every minute gets better. And then we all benefit from that. And that's where, but in some ways that iterative approach just can't be emphasized enough. Yes, it's siloed.

24:07Yes, it's been there. Yes, it's had quality issues. It's not had quality. Like you said, today, we don't just close the doors and time out because somebody made a mistake. we coach, we recover, we learn, we put controls in and we move on. And that's what's critical in this. You mentioned the data side, because data is the most critical thing with any system you're building. So there is the high bar there. But AI is also really good at working with unstructured data and making sense of things too. So you have this new weapon in a sense that can handle unstructured data versus like perfectly structured databases and things like that.

24:42Absolutely. If you think about what makes your best colleagues or your up and coming, it's their access to the data. Oftentimes we see the systems don't complement them. It's their willingness and it's their swivel chair and everything that they do to overcome those challenges. And so then now put on there the AI guardrails, get a practical starting point, start with high volume, low complexity processes. And a lot of times those have already been automated with something else. So now let's move into the mid complex processes or pick a mission critical process. That's very complex. The point is get started so you can train and improve the models quickly.

25:19And then it's just, it's representative. Disparate data exists. Appian as a company, we have what we call data fabric. We sit across the top of those disparate sources. We leave the data where it's at. We virtualize it in session. And that's what gives you the speed, the scale, and that's what makes AI incredible is access to all that, just like your best colleague, but they don't take breaks. They don't have quality errors. You set those guardrails and then put their ROI metrics around it so that the board can see transparently, the AI early adopters can see transparently, and that's how you just start to have those irritative wins and you just move on and repeat.

25:54Yeah. Let's go deeper on that. So you mentioned it earlier, like this build versus buy equation process, however you want to say it, is changing with the introduction of AI. And you get many enterprises that are being pushed to these all-in-one ecosystems like Microsoft, Copilot as an example or things like that, where it's everything Microsoft, which has benefits, but then there's also downsides as well. But I'm just curious, how do you think about those trade-offs? Because you have the kind of that clear strategy you just mentioned in terms of having that optionality while still having that security.