Get a demo
Select the type of demo you'd like to experience with Indico Data.
Select the type of demo you'd like to experience with Indico Data.
Watch Christopher M. Wells, Ph. D., Indico VP of Research and Development, and Michelle Gouveia, VP at Sandbox Insurtech Ventures, in episode 20 of Unstructured Unlocked with guest Jim DeMarco, Director of Insurance Strategy at Microsoft.
Michelle Gouveia:
Welcome to the next episode of Unstructured Unlocked. I’m co-host Michelle Govea, joined by co-host Chris Wells and our special guest today, Jim DeMarco. We are live at intro tech insides New York in a new room that’s got better sound for you. So welcome Jim to the podcast. Thank you. I think you two have known each other for a while through, through the work between Microsoft and Indico. Yeah. So love to understand the background on that a little bit and Jim a little bit on your background as
Christopher Wells:
Well. Yeah. Jim, you’re, as I recall, head of insurance strategy for Microsoft. Right. And that, that has been the context of most of our conversations, except occasionally sipping bourbon together and discussing all matters philosophical and otherwise yes. Is appropriate with bourbon does not <laugh>. That makes sense. But I, I may know a lot about you, but our, our audience doesn’t, so why don’t you tell them what’s up with you and what you’re into these days. Oh, alright. Well, thank you. Yes. So Jim DeMarco I’m the director of insurance strategy for Microsoft, which means I work in our worldwide financial services organization on how does Microsoft partner with and support the insurance industry all up. That would be working with our, you know, our biggest customers in the insurance business, but also with some important strategic partners across the industry as well to try to look at not just what’s now, which, you know, right.
But also what’s next and also a little bit about what’s after that because insurance companies as an industry are the backend of everything. You know, it’s, it’s the old saying, you know, the first thing you pay is the mortgage and the second thing you pay is the insurance. It sure is. And thank you Sandra Bullock <laugh> at the same time insurance touches everything and but it’s also known to be a fairly conservative business. So if we just focused on just sort of making sure the existing business works we would not actually help move the needle. Interestingly though, I think insurance is also one of the places where you see a lot of great innovation. This is where we actually saw data-driven decisioning That’s right. In the 1960s. Yep. you know, the first heavy use of mainframes came from this patch.
And we’re starting to see a similar sea change, I think now insurance industry as well, where we’re starting to see more digital first play, the idea of actually having insurance embedded in other human interaction mm-hmm. <Affirmative>. And so our ability to work with carriers and work strategic partners on how to move that needle for the industry and sort of make what’s next real Yeah. Is really kind of what we get the privilege of Microsoft to do. Cause we touch everybody. Yeah. But our ability to sort of help people is where we’re at, particularly excited to work with these guys. Cause while we are a really good platform company, you know, we make a lot of the platforms that sort of under rely a lot of how technology works, how you run things. We also don’t do everything ourselves. And one of the cool things is to get the opportunity to work with some of the real innovators in the marketplace around how do we change the nature of interaction, for example, which is one of the things Indigo data has been fantastic at.
Not for nothing. Microsoft’s really good at what we do, but we didn’t write the seminal white paper on generative ai. We are heavy investor and a good, and a good, a good partner for it, but there’s a really good opportunity for us to work with books like this. It’s very long-winded response to High Jim, but No, it’s, it’s good. Before we jump to one of my favorite topics, indigo, and of course, myself, <laugh> could we talk a little, that was all very, you know, grand and you’re doing big things and having big, but what’s the day-today like for Jim DeMarco? What are you doing to push all of that forward? It starts with waking up because the cat got me up a little bit too early Okay. In the morning. Yeah. yeah, no our day-today is, you have no idea <laugh>, but our day-today really is focused in on working directly with the, the largest insurance companies in the world and some of the, some of the sort of the other insurance companies out as well on understanding everything from where is their business going, right.
As to say, you know, what’s the nature of the larger business? But also then what are the sort of fundamental things that impact us? So we, we work really, really closely on things like climate change and how do we handle the role of climate changes, particularly timely given that we’re in New York right now. Yeah. Oh my goodness. The orange skies of New York right now. Blade Runner out there. Yeah. But with that said, we, we are actually looking at how do we use technology to shift the market to being ready for imminent risks and mm-hmm. <Affirmative> and how do we respond to imminent risks, whether, in this case fire, but certainly flood risk, weather risk and so forth. So that’s one of the things we get to sort of look at and work on, is how do we provide that shift in the industry from some sort of repair, replace to sort of more of the predict prevent kind of thing.
That’s really the day job we get to actually work on that. At the same time, we also work on things like public policy around cyber insurance and how do we sort of understand or work with the industry to help cyber’s a great example. Cause it’s totally a team sport, right? We have to make sure we’re, it’s good fighting the bad guys together, and how do we actually help people protect themselves. Yeah. And then the third thing, and last I to answer on the day job and the things that we really try to work on is how do we take and understand how the business is actually run? So part of my day job is really sitting down and actually saying, what is particularly generative AI’s been on people’s minds. How does that change the nature of what it means to be an underwriter, or the nature of it means to be a claim handler or the nature of the cfo. And that is conversation I have literally on a daily basis with customers. It’s
Michelle Gouveia:
Cool. So you’re sitting at kind of not, you’re, you’re helping develop and influence strategy. So at the high level, these big conversations mm-hmm. <Affirmative> at the industry level, and then also a little bit of stepping under the hood and saying, here’s, here’s what we can do to help you execute once you, once you’ve identified and developed that strategy.
Jim Demarco:
Yeah. It’s a fun job and I got it. You can’t have it <laugh> When do you sleep though? That’s
MG:
A big mandate,
JD:
Right? Whenever the cat Oh right, the cat. That’s right. But yeah, no limiting factor. It’s, it’s, it’s really kind of cool. We have a nice thing about Microsoft is we do reach everybody. That’s the good news. It’s also somewhat the bad news because when you do reach, everybody have no ability to do a lot of the, like the, the very, very specific. So while we talk about the general strategy, and we also work on how do we implement like execute a plan and how do you create a transformative moment for a business or for an industry that’s really cool. The arms and legs of deploying that change are things that, you know, thank God we have an army, right. So there’s that. The other side of that though, is also as broad a view as we have. We don’t have necessarily always the very specific view.
And so Microsoft as an organization, we’re a platform company. So our view is we can provide a lot of the technology and a lot of the sort of the reach capability and scale that is required to really, you know, be that fulcrum to to, to lever the world. But we don’t always have the ability to really get focused on here’s how I change this thing. Mm. Or here’s how do I drive something, I think an analogy that we’ve talked about together, and we’ll share with the, the folks here listening up home <laugh> is we’re kind of like the Lego set. We have the big green thing. Mm-Hmm. <affirmative>, you know, the bottom gets dusty on the floor. We’ve got the two by twos and the two by threes and the, the one by ones that always get stuck in the dishwasher.
Oh, I hate those things. Yeah. but we don’t make the Harry Potter Hogwarts train set. We don’t have the ability to sort of look at that. We work with our customers and we work with partners. One of the really cool things I like about working with these guys, <laugh>, because they’re really good at, he’s going at you, you the indico. I know me. Oh, no. Oh gosh. You are indigo. This is getting edited out. Yeah, for sure. <Laugh>. But one of the things we love to, we love is has the ability to sort of do the higher value add. Yeah. To do the thing about actually understanding what is the business process? How do I look at document ingestion? How do I look at the ability to create a flow through? And how do I create the edges that use some of the technologies?
Microsoft has some of the technologies that you have natively. Yeah. And sort of carry it to this solutioning that actually solves the problem. I have the privilege of being able to talk about what the problem is. I have a great opportunity to actually sort of help identify the parameters and constraints around how you look at a problem and sort of try to solve it. And I have a lot of platform technologies. We, we have a very strong relationship with a number of partners, indigo included, that actually then try to go solve those problems. And what’s really cool, one cool things about working with Indigo Chris in particular is these guys actually understand the problem really stupid Well too. We do. So there’s that. Yeah. And I, I would say, I’ll make, I’ll make a confession here. I, I’ve been at Indigo two and a half years when the sort of seeds of the Microsoft partnership were starting to get planted a year or so ago, I was like, Microsoft we’re like this trendy startup doing cool AI stuff, and they’re this dinosaur megalith.
Right? And I could not have been more wrong. Honestly, like the, the, the support, the collaborative nature of the relationship and also the, like, as you described, very clear boundaries between what Microsoft is, what Indigo is, and how that sort of building on top of the platform really adds value. I’ll also confess that the reason we understand the problem is that we’ve stubbed our toes on every corner of every piece of furniture that there is in taking ai to production in the enterprise and all of that toe stubbing we’ve baked into our platform. And you can build a lot of high value stuff through that. Yeah. I appreciate one of the things, and I’m not gonna sit here pitching Microsoft all day. <Laugh> one of the things that, that is really kind of cool are Satya Nadella, our CEO says, you don’t come to Microsoft to be cool.
You come to Microsoft to make other people cool <laugh>. And being an alpha geek myself, I’m pretty, pretty happy with that. And one of the coolest things here is, is that we actually like to partner up with people who actually can make things work and make sort of really, you know, how people achieve more. And, and this is a, a really cool thing in the insurance industry. Lots to change, lots of cool things that need to happen. And there’s a great opportunity for us to sort of change that help the industry as it transforms itself. So I’m very excited to, to talk also a little bit about that, cuz I get to play with that every day.
MG:
I love that you’re like, we’re gonna, Microsoft is gonna make anything in the insurance industry. Cool. Right. Those, those two words, they’re, it’s starting to get, people are starting to think it’s cool. Jim, we talk a lot about on, on this podcast about underwriting workflow, the claims workflow, the challenges there with data and automation and how do you improve those processes. I have a little bit of, of knowledge just from my time in the insurance carriers of, of what some of that might look like. There’s some differences that happen between carriers, but on, on the, the larger scale, a lot of those challenges are same. What do you find when you’re having these conversations with your partners? Is it similar challenges? Is it depending on where they are in their innovation or their digital transformation that they’re looking for different things? Is it very different as you’re talking to, to these partners?
JD:
Yeah. forgive an analogy cuz I think in them sometimes, but it, insurance companies situations seem to be like snowflakes. They’re, they’re all unique, but they basically follow eight patterns, right? Mm-Hmm. <affirmative> okay. There, there are certain patterns that you see in, in the market more generally. And so the answer to your question, I’m afraid show is, is frequently, it depends.
MG:
It’s my favorite answer. Ask
JD:
Me a question. Yeah. I never get a straight answer from anybody anymore. <Laugh>. Yeah. You, you put things on the internet, you know, what can I say? Okay. I’ll, at least you didn’t start your answer with as an AI language model. Yeah. <laugh>, yes. You know allow me to respond now. Yes. Hey, Chad. G P t. No. really seriously speaking what we find is that people look at technology to change things, but technology just accelerates things. We have to understand the business process that we’re trying to solve for. And every carrier seems to be in a somewhat unique position vis-a-vis their own journey. That said, I would also say that there are patterns of journeys that are very, very similar. For example, if I’m looking at say commercial property casualty claims mm-hmm. <Affirmative> mm-hmm. <Affirmative>, I know that if I have a fire claim, I’m gonna have certain things that are gonna come in.
I’m gonna have, you know, the certificates, the, the damage estimates, the building value estimates, the issues on what’s the commercial value impact and so forth and so on. And a lot of that’s gonna be a heck of a lot of documents that are gonna come in. And I need to understand what all that’s gonna be, and I’m gonna have some PORs schlep who has to sit there and look through all those documents and understand, you know, those documents and say, key up, this is what’s happening and so forth and so on. Mm-Hmm. <affirmative> and a lot of commercial property casualty is still done in that manner. Yep. some people have more sophisticated approaches to that. Some of them are using automated document ingestion. Some people are using commercial front end software like a CRM to actually sit there and do stuff. But at the end of the day, the pattern or the business practice is I need to figure out what went wrong, what’s the peril and whether or not I have coverage and how much it is.
Right. and that’s where we have to sort of ask the question, what problem are we trying to solve for? Am I trying to just solve for the problem of I want just a bunch of documents, or am I trying to solve the problem of I need to understand my liability quickly and at the least cost possible so that I can have the optimal result for client and firm. And when we start to decompose the business process, that’s when we bring in the technology. Okay. Yeah. I gave you a again, long-winded answers. I’m afraid that’s just my, my brand <laugh>, it’s less talking than me was gonna
MG:
Say it works for us.
JD:
Yeah. Jim, Jim’s coming back every week. Yeah. <laugh>. Oh God. Not for anybody who listens to this point.
MG:
No, but, but we’ve talked about that too, where if you automate a broken process Yeah. It, it stays broken. It’s just broke, it breaks faster. Right. I think is what, what Chris said. That’s
JD:
Right. Yeah. Peter Drucker, the, the the management guru. Yeah. Everybody knows him for culture eat strategy for breakfast. His best quote is that nothing is worse than improving that, which had never been done in the first place. <Laugh>. Yeah. No, we, so I talked about stubbing toes and getting to production with ai. Our plat, one of the things our platform allows you to do is sort of like memorialize your process in an AI assisted ai augmented and accelerated workflow. But we always caution people as we’re helping them get up and running. Like, is this, is this really the process that you want to memorialize forever and just let the robots run? Or, and we’ve started to build and we’ve started to build out capabilities for professional services to help people that have never really done change management on a back office, you know, process to, to think the right thoughts and to crystallize the right process. That’s actually one of the things that I do the most with my customers in terms of back to the what’s the day job? Yeah. My team, what we try to actually go do is just understand the current and intentional process. Yeah. for things because, you know, given the quotes, I guess Einstein was asking, Einstein says, if I now to solve a problem, he’s pulling
MG:
These off though.
JD:
It is someone fact checking this. Yeah, yeah. Real in real time. Somewhere around here, banner, somewhere in here, there’s generative AI going, here’s your next quote, Jim. But you know, Einstein said, if I had an hour solve a problem, I’d spend 55 minutes thinking about the problem. Five minutes on the solution. Yeah. And the reason for that was you know, his, his, his rationale is if you can decompose the problem to actually what you’re trying to solve for Yeah. You find that the solution actually comes if you just start right into solving the problem. Yeah. you end up solving a piece part and not actually getting to the end. Yeah. So when you guys are sitting talking with your customers and they’re going, Hey, I really want to handle this really cool thing about using tech to actually make it so I don’t have to have an army of people sitting their hand coding responses in moment. It’s actually the part of the challenge really is, well, that’s a piece of the, the overall problem. The problem, the KPI I’m going for is I need to get a fast decision that’s accurate. Okay. Now let’s look at how do we solve it. We
MG:
Just talked about this on a, on another episode with, with brandy from, from Indico where she was talking about when she’s having these conversations with, with the customers and launching pilots or, or starting the engagement of, well, the customer says, this is what we wanna solve for, this is what we wanna measure. And part of Brandy’s role is having to roll it back and say, but to get there, you’re really trying to solve for X. Right? Right. And you’ll eventually get to that metric. But what’s important is, is solving the smaller piece or the Sprite piece, I guess. Yeah, yeah,
JD:
Exactly. The decomposition of the problem into the, you want into its piece parts and then the, the providing the right flow. Yeah. It’s just all that math we did back when we were kids. Turns out we were kids. Okay. <laugh>, sorry. When you were a kid, I, you know, maybe for me it was a long time ago. We were still using, you know, we were using stone slates back then, but yeah. Yes. I didn’t wanna say that because I couldn’t remember the word itself. No, no. But seriously, that construct of making a big problem into a smaller one and then understanding what’s the intentional flow, that is the essence of transformation in many respects.
MG:
Incremental improvement and addition
JD:
To,
MG:
To get to, to the ha you know, the pie in the sky state.
JD:
Yeah. And, and, and at, at the same time, sorry, as an aside from what I do for Microsoft, I also actually aside hustle, I teach graduate school and I teach digital transformation. Bless your heart. Oh, yeah. It’s my poor students. Many. I wasn’t sure where I was going. Many listening to this. Sorry. Yeah, no, I like, you’re allowed me to show you my company <laugh>. Yeah. No no. But what, what I see is we, we, we work with, we work on digital transformation for large and small companies as part of this course mm-hmm. <Affirmative>. And what I see, and I continually get the response, people think, oh, digital transformation is I’m bringing all these new technologies like stop mm-hmm. <Affirmative>, right. Digital transformation starts with where is your business? What are the things that are changing that business? And then how do you look at yourself in the future and look at the business intent. It is a strategy around what am I trying to change for my business, for my industry? And then come at how, how I leverage that technology. And that for me is the essence of how I get there. And then if you’re gonna do it, then you have a roadmap, and that’s where you pull in small piece here, larger piece there, and so forth. And then sort of create that flywheel effect of change. That’s right.
MG:
I think a lot of people too one, one of the, the criticisms of the insurance industry right? Is that it’s so slow moving Right. To innovate, to change. And I think, Jim, the point you just made is, is a perfect one in that you have to do it incrementally because these insurance carriers are enterprises that have been around for many, many, many years with their own legacy systems or technologies, that it takes time to deload all of the information, the data that’s there, to get it into one a place that can be part of a digital transformation, but even to move it oh, in a way that, that that data stays, you know structured the way it is or has been used in the past. A lot of different systems are siloed. They need to start talking to each other again. I think once, once startups are, you know, people understand that it becomes more, a much more, to your point, digestible challenge to say, this strategy will take us five to 10 years to execute, but if we start doing this now we’ll start to see constant stream of reward.
Exactly. Yeah. I
JD:
Hate, I, I’m sorry for this dad joke, but I think the emphasis has been on the wrong slab in digital transformation and that transformation is the thing, right? The digital is an implementation deal. Yeah. Detail, frankly. Yeah. Tech technology is an accelerant. Exactly. and, and while there are certain technologies that are such accelerants and touch so many things that they can be considered transformative in and of themselves mm-hmm. <Affirmative>, at the end of the day, it’s still solving for a human or a business process. That’s right. And that from our perspective means really start, come back to what are we trying to do? Where are we trying to go? Back to your point, Michelle, and like a lot of insurance companies, insurance has a sort of a, please forgive me a little philosophical here, but insurance has historically been built around understanding specific risks mm-hmm.
<Affirmative>. Yeah. And customers were categorized and bucketed into risks. That’s how we would be risk pool, right? That’s the whole point of the exercise in some respects. But as a result, we have a very product driven, risk driven model where you are, as a customer to me, you, you would be a set of risks associated with it until I tell you a product that is, he’s this package of risks. If you wanted to buy another product from me, you’re a second customer to me in many respects. Right. Because that product has its own sort of bucket analysis of risks and so forth and so on. So one of the fundamental challenges that insurance companies have run into is they, they address it as we need to bundle bundling is, is, is a symptom of the, of the broader problems. I need to actually understand how my customers and what they’re trying to do.
Right. And then if I say the talk of going customer-centric as an, as an industry requires us to sort of rethink fundamentally what is our operating model and our operating intent. And that is a change in not just process as a transformation mindset as a business. Mm-Hmm. <affirmative>. Yeah. I had this great conversation with an insurance company a number of years ago who actually sitting in on a conversation with board of one of these companies, <laugh>, and they’re, they’re talking back and forth about all these cool things they wanna do with technology. And I, I kind of raised my hand in the middle of the meeting and, and said, so just outta curiosity, you guys have a customer model. Chief marketing officer leaned forward, looks at CIO square in the eye and said, I told you so <laugh>, it was, it was really cool.
It was, it was, it was kinda like being at Altima at that point. It was like, eh, it was a big fight. But it was really, really cool to understand was there’s a fundamental understanding that customers don’t belong to products. Products are things that support customers. And when Right. If somebody comes to an insurance company, they’re asking for protection or advice or something about their own personal lifestyle, their business or whatnot. Yeah. Mm-hmm. <Affirmative> and the insurance companies themselves have to sort of really realize that those become, that becomes the center of the process. The process is a human or a business trying to achieve something. Yeah. Rather than somebody that’s coming to look to buy a product off of me. And that’s a fundamental seed change in terms of how we look at our systems, our processes, and our integrations with others.
And if I’m sending you a bunch of documents and I’m saying, here’s my stuff, I don’t under say, oh, fit that document into my processes. No, these guys actually trying to do something. That’s where you see, I think the difference between good and great in five years from now in the industry. Are you seeing, sorry, are you seeing carriers have to, as they’re making this transition, have to reeducate the customer? Because this is not the way the customer has come up to the counter and asked for the product historically either. It’s not, but there’s a great analogy in, in to all this. About 25 years ago, telco ministry did exactly the same thing. Okay. Interesting. Cuz everybody, you just subscribe to a plan, remember that, and then we started offering things like bundles, and then you could actually do multiple things with your cell phone, and then you had to like, do more and more services.
But this turns out you started thinking less about like, what are the services I’m putting out to the market and how do I make profit on those services? Yeah. But now how do I look at my customers? Yeah. Okay. So there is actually the customer I think insurance companies more generally, we have a mindset in the industry that we sort of understand risk really well, and no question the insurance industry does, but understanding that customers don’t think in terms of risk. They think in terms of how do I run my bakery? How do I make sure that I can get my kids to school? How can I plan for my kids’ college education? Or how do I actually drive a car? You know, how can I actually make sure I can go drive a car? Those are what the customers, the humans are actually trying to achieve. I guess. I’ve never looked at actuarial tables before I went to buy car insurance. That’s a good point. <Laugh>, it was that, you know,
MG:
They also keep them very secret. So you couldn’t, you wouldn’t find ’em even if you looked yeah,
JD:
Congratulations, you are in this box that appropriate.
MG:
But I think too, that’s probably why there’s so much conversation and interest in the embedded insurance model, right? Because you’re, you’re bringing those insurance products to the point of where the, where the customer is in a, in a lot of times, and that’s, they’re, they’re, they’re looking for a product. You’re meeting them where they are, even if they may not even realize they need the product, but it’s there. There’s that synergy of, oh, I’m buying this. It makes sense to have this. And I think that’s why you’re seeing such, such high demand for, for that. Cause it’s a lot of lifestyle products. You know, I’ve seen a lot of insurance that’s like, oh, you’re going on a snowboarding trip, you should buy, you know, accident insurance, <laugh>. Okay,
JD:
Amazing. Plus travel delay plus Uber insurance. Plus, plus. Yeah, exactly. To your point, yeah. That’s, that is a sea change. Yeah. But that is also a sea change where insurance is supporting a human doing something as opposed to, oh, wait a minute, I need to go to the insurance to go do this. Exactly. Yeah. There, there will always be a world where I need to have a financial backstop if my building burns to the ground. Mm-Hmm. <affirmative> mm-hmm. <Affirmative>. That is true. Because at some point or another, if I don’t have insurance and I have a house and all my money’s in my house and my house burns down, I’m kind of in trouble. That is very true. But when we see the, when we see now people traveling, or we start to see the idea of embedding insurance in a product with like, you know, warranty embedded mm-hmm.
<Affirmative>. Yeah. We, we can skip the details on warranty insurance, but that the construct still is the same. It is something that supports another human transaction. And that I think is where we have to really now think insurance is a slightly different thing, right? Yeah. It’s now much smaller. It’s much more rapid it to be something that fits into something that a human can go do. There’s some really, really cool stuff out there like travel delay insurance. Mm-Hmm. <affirmative>, I used to you travel delay insurance. Yes. Yes. Did you get your $50 for your, your dinner?
MG:
$10? It is parametric, so yes. It’s, it was triggered just by the data that said the train is late here is here is your reimbursement for, for that
JD:
Ticket. And look at Michelle’s face lighting. It’s about parametric insurance.
MG:
Insurance. Gee, I know. Yeah,
JD:
I know. That was great for the, for the 17 people who know what parametric insurance is. <Laugh>. That’s awesome. That is. But that’s also I think the right digital model in a lot of respects, because what it does is it gets away from this construct of I need to think about 400 years of floss history mm-hmm. <Affirmative>. Yeah. Yeah. Down to it is a small, simple transaction that supports a human doing something you getting on a train going someplace. Yeah. Right. And that, that is now, that’s how we talk about transformation in the business. Now, when we talk about that, there’s also, that means we have to, there’s nobody who could underwrite that manually. Parametric insurance. You needed to have the parameter set. Yeah.
MG:
You need to have the data feed. It needs to be automatic. Yeah.
JD:
Right. And, and that means that’s where tech comes in. Mm-Hmm. <affirmative>. And I think that’s exactly to our e Exactly. I guess the point of your, of your podcast tech is now supporting a completely different process now it’s supporting for sure. Plugging in so that you can, you can go someplace mm-hmm. <Affirmative>, right. Or, you know we, we, we talked with somebody about trying to put together a plan for Coachella Insurance. Right. You know, all of the different things that, that go into my, my, my experience
MG:
Space light up please.
JD:
Yeah. About Coachella. Coachella, let’s just say that’s not what I do on my off time <laugh>. Right. The cat won’t let you go to Coachella pretty much you know, some stodgy old professor. But at the same time the construct becomes now looking at how do I slipstream that into a human process. Mm-Hmm. <affirmative>, there’s some people who are absolutely phenomenal at this in the industry, and surprisingly not always the startups you know, oh, InsureTechs are going to, these are actually some of the biggest carriers. You’re doing some of the really coolest stuff, travel, delay insurance. Mm-Hmm. Was that, that that wasn’t, was one of the largest reinsurers I think you would’ve invented in that. Huh. and that’s cool, you know. Yeah. I, you know but it, it takes a mindset shift mm-hmm. <Affirmative>, and, and it really comes down to finding the right business process and then asking the question, how do I put the technology into it?
Yep. what that means for us tech players, for guys like you and, and me Yeah. Is it means we have to be super quick. We have to have stuff that’s easily scaled and easily deployed. Right. and then we need to have stuff where it can be just, I, I learned from this and I grow it, and I grow the next, and I grow the next, which is again, you know, we start looking at back to sort of what’s the core bread and butter of like, how do I use AI and insurance? Yeah. Yeah. Then we have to understand that it has to be understandable, rapid, easy to understand, easy to use, easy to scale. And then I, I don’t need to train an army of data scientist in how to use this stupid thing. Yeah. Right. And, and not the data science isn’t a wonderful experience I got. That’s great. I got a daughter’s a data scientist. But what they’re shifting more and more toward now is understanding how, you know, really good data science with people who can understand what’s a human process so they can actually configure how that works. Right. Yeah.
MG:
You touched on, on the major topic, I mean, here at the conference it’s been, it’s been keynote speakers and, and panels on generative ai. It’s the big topic of, of the conference the
JD:
Flavor of the day. <Laugh> the flavor of the day. Yes. Flavor of probably the next five to 10 years. Yes, exactly. This is a game changer. We,
MG:
We’ve talked about it a little bit on the podcast, but would love to get to get your thoughts, Jim, on, on how do, how do you go about deploying those capabilities? I mean, ai, we’ve talked about AI’s been been around yeah. For a while. Like companies have been using it, but now there’s all of this attention on it because individuals can, can use it, can access it in those capabilities. And how, how do they get to do that? Right? What are, what are the concerns? What are the, the things they have to be worried about? What do carriers need to think about when, when they’re looking to deploy solutions with these capabilities and both of your thoughts on it? I mean, you guys are much more experts in that space than I
JD:
Am. Well, I, I think I’ll actually deflect that question and ask Jim a related question before I give my answer, which is we’ve, we’ve heard from a number of carriers like here and in recent conversations saying things to us like, oh, GPTs here we don’t. Why would we need a solution like yours? And I wonder if you’ve, part of that I think is part of it I think is hype. Part of it I think is the usual sort of business mindset that here’s a technology, we’re gonna technology all the things. And then part of it I think is just wishful thinking. But I would love to know if you’ve had similar conversations and how you answer that question. I’ve, I’ve heard everything from Skynet is now self-aware, <laugh> this, this AI stuff’s gonna kill us all, which I think is perhaps a bit extreme that
MG:
Seems be making headlines it us.
JD:
Yeah. Come with me if you want to live Jim. Yes. It’s a bit on one edge. On the other edge. I have seen the, this stuff’s never gonna work. No. and the, the sort of, the, the other piece of it I think we’ve seen is use this for what <laugh> and, and it’s actually the use this for what is probably the most interesting of those three responses. Cuz this is kind of not self-aware. This is, this is narrow way. I, we can talk deep, deep details about you know, the construct. But it, it is taking the technology, generative AI is taking you know, it, it’s taking drudgery out of work. Right? Absolutely. It’s absolutely, it’s got the ability to reduce the use of rope task. Yeah. Great. Back to our earlier sort of philosophical conversation about how does all this stuff work, right?
We’re still thinking the problem is what’s the business problem or technical human problem we’re trying to solve. Right. So, so back to sort of what we are seeing and hearing in that space then is we’re, had the questions that come to our, come to me from customers and quite literally on a daily basis now is, what can I use this tech for? Mm-Hmm. <affirmative>. Okay. And my response back is, it varies, but my response back for it depends. Yeah. It depends. Of course. I, I didn’t want to give the consultant’s answer, but thank, thanks Michelle. My, yeah. Appreciate, you know, you’re teaming up on me. This is not how this podcast usually goes. I, I have my pen. I’m gonna write down the number of times we say depends in the conversation. But the, as it as it comes out, the, the response I give is we need to actually understand the business processes we’re coming at mm-hmm.
<Affirmative> and then understand from there what I can use the technology to, to achieve. So I actually sit down and sit with customers. I wanna talk about generative AI is there’s sort of three basic, three basic pillars to the, to the, to the response pillar. One is the use cases. What am I trying to do? Pillar two is a technology play, cuz this is tech. Yeah. You have to deploy it. You also have to deploy it securely scale. Both, both vertical and horizontal. So vertical as in I need to use a lot more of it sometimes and horizontal is, and I use it for more things. Yeah. so I have that piece. I mentioned security. I’m gonna come back to that three or four more times, but do as many take marks on security as I can. On, on it depends <laugh> because it’s really, really important.
Yeah. yeah. You know, everything from regulatory, we’re working with customer data mm-hmm. <Affirmative>, we’re working with you know, sometimes really super sensitive data. Yeah. P I I P h I you know, P x I, if you will mm-hmm. <Affirmative>. and, and there there’s elements of all of that. So the technology deployment model has to be done. Right. and just saying, oh, I’m just gonna go use chat g PT to do it. <Laugh> Microsoft is very well known. We’re a huge investor in open ai. But at the same time, we need to have industrial grade scale. If we’re an insurance company, we need to have highly compliant approach to things if we’re we’re the thing and a system that’s flexible and nimble enough to change as the regulatory landscape changes. Right. Right. Yeah. And that’s actually where the third leg of the stool comes in.
So use cases, technology and governance. Yes. Exactly. To your point, that third leg of the stool is doing it right. Then doing it right is a combination of responsible ai. It’s a com. It goes with all of your compliance, your risk, your risk modeling, your compliance modeling, and how you actually operate as a business. There’s great technology, whether we talk about generative AI or other cognitive ai, which is the other place where we see a heck of a lot of work. Mm-Hmm. <affirmative> coming in. There’s a lot of things that you can do with that technology that to, to quote the guy who’s like a few layers above me in Microsoft, it can be a tool or it can be a weapon. Yeah. And has to be done in a manner that is responsible. And that’s when we sort of, when I work with carriers, I say, this are the three legs of your stools.
You need your list of use cases. You need to, you need the technology play to be done. Right. Get your data house in order, get your integration pieces in order, get your business processes laid out. Right. Make sure you’re doing secure scale compute. And then incredibly importantly, and I know a lot of, a lot of companies and, and a number of insurance carriers miss this third leg of the stool, which is governed properly, governed ethically have have a real, at the board level or at the C level, have a governance committee for how you use technology in this way. I would be remiss if I didn’t point out to the listener that very little of what you just said is open AI specific. Like, no, this is, this is true of any technology you buy. I and I, I also and know, no offense to my insure insurance colleagues out there, but governance I think has long been a struggle in insurance companies where every actuary has a model they run on their desktop with their own access database. No one
MG:
Know about it. We’re seeing a lot of movement now on, on the entrepreneurial side of companies coming forth with these governance and compliance solutions of here’s how we can audit, audit your, your models. We can have a track record that you have reviewed these, that there’s no bias in your models. And to where’s that data coming from? We have the protections in place to, to secure that data. We have protocols and, and procedures if we need to eliminate that data from, from a model because of, of, you know, a state has said you can no longer use that in your underwriting or your pricing or, interesting. Okay. We still have to do that regulatory episode. We
JD:
Do have to do the regulatory episode. It’s gonna be a great one. We just, we
MG:
Need, there’s anyone that does rates and filings work out there that’s listening that wants to join. Please do. Yeah. it comes up every time we talk about this. It really does.
JD:
<Laugh> Well, it’s interesting. So I’m gonna jump in on your regulator thing because actually one of the things we’ve been advocating as a business is to actually establish well formed regulations, well-informed regulations mm-hmm. <Affirmative>. Yep. Around this we have within Microsoft, we have most, is one of our real distinct advantages or differences in the marketplace. We actually have a committee, an organization that actually sits there and looks at the use of ai, ours and our partners. Nice. And actually we create our own sort of self-regulating model around whether or not we will use a piece of ai, whether it’s generative, cognitive Yeah. Or deductive you know, machine learning. We actually go through and review project by project. Is this a project that could be you know, what’s its intended use? What’s the, you know, is is it fit for purpose?
But then also what are the potential unintended consequences? Yeah. I worked with a partner, absolutely phenomenal technology that was really, really cool. It actually has healthcare outcomes that could be very, very cool and very positive, both from an insurance perspective, but also just from a human surviving perspective. Yeah, that’s good. Very cool piece of technology. But in certain circumstances it could be used to discriminate against people. Okay. And so we said we actually, we have actually stopped our partnership with them until they built an entire framework and guidance guide, guide rails around the responsible use of it to make sure that it could not be used in an inappropriate manner. And that’s part of how you do this. If it’s done right, insurance companies need this for compliance purposes, but they also need to do it just cause it’s the right thing to do.
Yeah. Mm-hmm. <Affirmative>. And that’s a place where we really want to work with customers. Not to pitch our partnership or nothing, but you actually have some really cool ideas around actually, because generative AI is part of a challenge in this regenerative AI and of his nature. It’s a bit of a black box in terms of the output. Oh, it’s the blackest of black boxes. But there are places where we can come in and actually provide traceability by actually sort of logging, handling outcomes, the flow through process. And that’s actually some really cool stuff. So would you mind sort of talking through that a little bit? Cause I think that’s cool. I’ll give a funny anecdote first and then maybe I’ll talk a little bit about what we’re building and what’s coming out, you know, even in the next few weeks. The funny anecdote, and I think this goes back to what you were saying, responsible AI is a super nuanced thing.
And I’m, I’m glad that smart people like you all are, are working on it. Cuz I think most people would put some pretty draconian and not very thoughtful rules in place. But even, even even the rules that maybe are in place today are, could be a little more nuanced. So I was, you get in trouble for this. No, I won’t get in trouble with this. This will be fine. <Laugh> at my next company, <laugh> I was I was trying to generate some synthetic data you know, rare data fields that were found in only a few documents. And someone had used to train a model. So I said, well, I’ve got this big generative black box. Let me put the few examples I have in and get out a few that I can use. And then, you know, loop that back in.
Retrain the model, you get this like virtuous cycle, right. At least in theory. And there’s this one field and these were all like commercial real estate documents. There’s this one field that I, that I kept, like it just aired out. And like, I wasn’t, it was bad software practice. I wasn’t logging the messages very well. So I came in and I looked, I just sent one through and it was like, you violated our terms. And it turns out this was about fair use of a building. And the, the, like, one example I had talked about brothels and sex workers. And so I was trying to do something very ethical, which, you know, help someone get a machine learning model that wasn’t discriminating against anyone. And I got blocked. Because the filter, you know, it was, it was a very hard filter, right? Like the context of what I was doing was not part of making the decision about whether I should do it or not.
And so I, there’s work to do there. I would say it’s, it’s kind of cool. So, and, and I’m, I’m, I’m loath to get into product details and as a strategy guy for Microsoft, I’ll probably get the product details wrong anyways, but we’ll our next company together, Jim <laugh>, the other side, I’ll be here, Michelle. Appreciate that. We’re gonna need some investing quickly. But one, one of the things that Mike, that that we do, people actually ask about sort of why, you know, what is Microsoft with the open AI stack? Yeah. And Yeah. Yeah. One of the things that is inherent in what we do is our systems, we actually track input output on the use of generative ai. Yeah. And we do not record we actually have something that’s just this quiet lister on the output of any trained model. So, you know, model response.
Right. You know the tech better than I do. If certain things start to come out on, on the use of any given model, we actually, the system auto logs it and flags it. Nice. So it recording? It’s recording. Okay. Pick up where you left off or something. In that neighborhood. You’re talking about the logging of outputs? Yes. Yeah. So one of the things how they, the Azure, Azure Azure Open AI stack. Yes. It kind of sort of works. And for those of you who are listening in, please don’t hold me to it, <laugh>. I’ll get the text disc disclaimer, I’ll get the text slightly wrong. But the general gist of it is this. Yeah. When, when you’re dealing with the use of any algorithm with cognitive or with generative ai, generative AI in particular, we actually, the system in the backend is just sort of letting it go, letting things happen.
But if some, but it is holding back, we do not, in our contract, we will show that we, we will sometimes stop and flag and auto record something as an output, even though it is not our data. Mm. Because so people who use their data, they, they’re doing their thing. Yeah. We will actually do a hold if there’s an inappropriate use flag hit. Okay. So the idea of talking about sex workers may trigger that it’ll get logged. Sorry, I’m in Use you last. Yeah. And it’ll get logged, flagged, and the human will actually have to look at it. Wow. Okay. Now, if a customer decides they don’t want to have that logging, they can turn it off. But the only way they can turn it off is if they have an ethical AI program that does it themselves. So they actually have to have they, to prove this, they have to prove they actually have their own structure in place for us to do it.
And, and, and the point of the exercise here is that this tool is incredibly powerful. These tools help develop the very first of these tools. These tools are super, super powerful, but they can be used for inappropriate use, even unintentionally. Mm-Hmm. <affirmative>. And so our ability to sort of make sure we’re doing the right thing is we pride ourselves in a position where we have to self-regulate and self-police. That said, back to your thing, should we do something on, on, on regulation? I think the answer is yes, because we need to be working together with whether it’s, you know, in, in the case of insurance, any IIC or, or, or I a I s but also with the governments themselves on these things. I know there’s the, the, the US government’s working on this. EU is working on this. UK government’s trying to work on how do we create the right standards and frameworks appropriately.
Yeah. And that’s a place where we think it’s a great opportunity for public private partnership as well. Fascinating. So, on that note, let me circle back to the original question you answered me, which is about where is this going within Indigo Indigo’s history, we were very early in, in generative models, and then we realized that the discriminative models at the time were more powerful models that can’t generate stuff, but they’re good at finding stuff, right? Right. So we built a product on top of that. And that’s nice because they don’t, the discriminative models, you know the name’s discriminative, like they don’t misbehave. Yeah. Right. Like they’re just finding what’s in your document. If you use it for bad things downstream, well that’s, that’s on you. Like we just, again, bad process. We just helped you get there quicker. Right. so we’ve taken that mindset.
Everything we’ve learned about, you know, comparatively discriminative model, they’re a bit of a gray box. They’re not quite, you know, coal black. Right? Like a, like a large language model. We’ve taken that mindset to the large language model space and we’ve said, okay, one of the things you have to do to build a scalable machine learning model is you have to label data. So let’s use GPT four zero shot and just, you know, help make it turn into a labeling engine. Like, do my labeling for me. I’ll be QA instead of the highlighter on the page. And one of the things, one of the things that changes there is you actually have to, you have to take its answers. It says, I found this, but it doesn’t, it doesn’t know where in the string the word started or where it ended. That’s just not the kind of thing it does, right?
Mm-Hmm. <affirmative>. So you have to go back and find it. And so it’s a combination of prompt engineering, good old fashioned software engineering, and also just some heuristics to say, okay, we, we think we found it here, there, the model thinks it found it here, we can find it there. It’s not a hallucination. Okay, good. Push that into the label. Let the humor review it. Right? Now you have traceability. Now you have traceability. Right. Which is, which is huge, right? I, I came from financial services and data governance and like the flow of why did I invest in this based on what I saw in the markets. It’s like you have to have that, you have to be able to cite it chapter and verse. Cuz someday an auditor’s coming looking. Right? It’s not a matter of if, it’s a matter of when. Yeah.
And also decision you in the insurance space to, to tracking to why did I underwrite this way? Why did I actually determine exactly what determination did I make about a claim? You know, all of those different elements, those are all required traceability. So there is a compliance element regardless of financial services, industry sector that you have to get to. And this has been a fundamental challenge as people start to look at this like, oh, we’re gonna do with it. Well, the cool thing is you can do an awful lot, but you can, when you actually add in the ability to drive traceability of large language models, now we can think about how do we simplify tons of processes and still stay compliant. Absolutely. And it gets even scarier when you leave the sort of realm of like, help me do discriminative stuff, but faster to help me truly generate something novel.
Like, here’s a summary of the policy coverages that I’ve written so that I can compare with like the first notice of loss that came in, right? Mm-Hmm. <affirmative>. And you know, with a little bit of work, you can actually do not just summarization, but you can do grounded summarization, right. And have the model show its work, right? Like, I don’t know whether I think these large language models can reason, but they can, they can act like their reasoning. And if, if you keep that, if you have the right mental model for how that works you know, you can sort of open up the box and you can see what, what’s going on and why a decision was made and you know, why this word followed that word. Right? And now you can simply doing this, you can simplify the business process. Hundred percent. You can squeeze it down.
So now it fits in with embedded insurance, with embedded finance. Yeah. Because now you can actually say, I can, for the easy stuff, yes, I can have a fully traceable, fully understandable, relatively clean cut to the you know, to my my target loss ratio approach to my business where I’m not just taking my best guess, but I actually based upon the appropriate parameters. And then when there’s stuff that’s hard Yeah. Then I can actually, that’s where I bring in my humans. That’s where I actually start doing the more you elevated their role. Yeah. Yeah. This is, it’s only making human jobs better. Yeah. At the end of the day, I love it that people say, you know, Skynet is, you know, sky Skynet is self-aware, is is kind of funny. The, oh by the way, we’re gonna eliminate 10,000 jobs doing this. We’re just gonna make 10,000 jobs a whole lot easier.
That’s right. Yeah. And enable people to do more high value add work and a lot less drudgery, because now I don’t have to have those people go sift through the document to find Yeah. You know, is there a family history of heart attack? Yes. No. When I get back the answer yes, no. Right. And by the way, the turnover on those jobs is miserable. Right. Because nobody wants to do that forever. Yeah, yeah, yeah. Exactly. And now I can let people be a better, get more expert at what they do. Yes. and there is a virtuous cycle in all of this if we do it right. But it still all boils back. You have to have the tech right. You gotta get the use cases right. You gotta understand what problems I’m trying to solve. And then if we don’t do this, if we don’t do this in an ethical, responsible manner, that’s right.
We are not long for this world. That’s right. And I would say Microsoft and, and Azure open AI are at the forefront of that. At least one man, just one man’s opinion. But I do spend my day thinking about this. Well, from, from, you know, as my old mentor used to say, from your lips to God’s ears, <laugh>. There you go. Well, we have covered a massive amount of, of ground today. Any last words, Jim, for the audience? Well, so thank you for, for suffering through a conversation that involved me, but also I, I have to say, I’m, I’m thrilled about the direction and the ability that we can take things. We have. This really is because there are so many different places we can touch now with, with generative ai because cognitive AI has finally become a mature product. Yes.
This product set as well. We now have the ability to really rethink the human process and simplify, simplify, simplify. Yeah. That, for me is super exciting. Working with great partners who actually, you know, can help push the thinking and also make it really work and bring it to life is fantastic. Yeah. I’m really excited. I I am, I’m an optimist generally, but I’m super optimistic about this. Same, thank you. Well, thank you. Thank you, Jim. Thank you. This has been another episode of Unstructured Unlocked. I’m co-host Chris Wells. I’m a co-host Michelle Govea, and our guest today has been Jim DeMarco, who does all things strategy for insurance companies at Microsoft. One last. Thanks, Jim. Thank you. Thank you. Take care.
Check out the full Unstructured Unlocked podcast on your favorite platform, including:
