← AWS Executive Insights
AWS Executive Insights
How Startup Parsyl is Revolutionizing the Insurance Industry with Generative AI
Loading summary
Transcript47 lines
- [00:00]
Jake Burns
Foreign.
- [00:04]
Podcast Host
Welcome to Conversations with Leaders, an AWS podcast focused on personal lessons of leadership, culture and technology from business leaders across the globe. Enjoy today's conversation.
- [00:18]
Jake Burns
Welcome to today's episode. I'm Jake Burns. I'm an enterprise strategist with aws. And today I'm joined by Mike Linton. He's the co founder and CTO of Parcel. Mike, welcome to the podcast.
- [00:29]
Mike Linton
Thanks Jake, appreciate you having me.
- [00:30]
Jake Burns
So can you tell us a little bit about Parcel, what you do and what is your role there?
- [00:34]
Mike Linton
Yeah. So Parcel is a data powered insurer of essential supply chains. Our overall goal is to work with customers to navigate the modern risks of supply chain that are in this rapidly changing world and guide them towards more sustainable outcomes. My role as CTO is I oversee the technology strategy and direction for our data powered insurance business.
- [01:01]
Jake Burns
That's great. So I heard a quote of yours that Parcel's mission is to reduce waste in the supply chain. And I think that's such an amazing mission to have. Can you tell us a little bit about how you're actually doing this?
- [01:14]
Mike Linton
Yeah, it's a very interesting thing where we, you know, significant amount of goods as they get transported around the globe arrive in some type of degraded state. And there's a lot of reasons why that is. And when we started Parcel seven, eight years ago, we were actually looking at this issue within global health and within vaccine delivery because it's very pronounced and it's even more pronounced in low income and low infrastructure countries. And when we originally set out, there wasn't much data collected on it. And so what we wanted to do is to do something about the problem. But there's not much you can do about the problem if you can't quantify it, pinpoint it, obviously that's where you use data. And so we went out to collect as much data as we could on these sensitive supply chains. And over many years and all kinds of evolutions, we've evolved into this insurance company. We did not set out to start an insurance company. None of us have backgrounds. Insurance company, I should say the founders. We now have a team of great insurance professionals. But the mission has largely remained the same. Where we are looking for ways to help quantify this risk and ultimately reduce it. And so when we actually can collect more data and then ultimately now as an insurance company, we can price the risks, that allows us to actually work with customers to do something about it.
- [02:37]
Jake Burns
Yeah. So I imagine that better understanding the risk will allow you to offer better insurance rates for your customers. So how does that work.
- [02:45]
Mike Linton
Yeah. So traditionally when a customer is going out and needing to get various types of supply chain coverages, whether it's their kind of cargo coverages or delay risks or product recall, political instability, all these different types of things that can cause issues throughout their supply chain when they go to one of the traditional insurers out there, there's a fairly limited amount of data that is actually used to understand what that risk is. And so we work with a significant more data, whether that's from the customer. And I can go in a little bit more detail about how we're actually engaging with the customer's data. But when we can actually get a better view of that risk, yes, we can offer a better rate. Because what happens in the absence of understanding in the insurance markets generally, if you can't view that risk, you're probably going to price it a little bit higher. Like if you can't really quantify it, you're probably going to go a little bit higher. And that's just traditionally what happens in insurance markets. And it's not because people want to price it higher, it's just that if you don't have a good view of the risk, that's kind of what happens. And so when we can pinpoint it, yes, there's a lot of times where we can actually give a better rate, or sometimes that comes in the form of maybe better coverage or deductibles or other types of incentives. But what we really want to do is we want to provide transparency to the customer about where their risk lie. Because that's where the real key comes in, is we want to enable customers to actually do something about it. So even if they are maybe priced high, we want to say, yeah, it's because maybe you're using this particular trade route or port or partner or packaging material or all kinds of different factors. Well, if they can see that, then they can actually start doing something about it. Maybe it's something they can do immediately in this next policy period, or maybe it's a longer term thing, an investment that they need to do. And as a commercial insurer, we really want to work with customers over a long period of time. So we will really work with them, especially if they're making those investments throughout their supply chain.
- [04:44]
Jake Burns
That's interesting. So it sounds like you're actually shaping your customers behavior by providing this transparency to them using their data and also other data that you have, if I'm understanding you correctly.
- [04:56]
Mike Linton
Yeah, that's exactly what we want to do. Especially in the type of insurance that we're in, we're in this unique position where our incentives are very closely aligned with the customers. Customers don't want shipment of their goods to go bad. That has all kinds of other risks aside from just an insurance claim. We don't want an insurance claim. Obviously that's bad financially for us. And so when they do better, we do better. Right. So it's the alignment of incentives is the real key here. And so that's why we make all these investments in data acquisition, technology, automation, all these different things. So that way customers can get that transparency into their risk and do something about it.
- [05:38]
Jake Burns
Right. And I imagine there's so many benefits from doing this. Right. From the sustainability aspect of being more efficient with shipping to getting their products shipped more quickly, serving their customers more quickly. I mean essentially you're finding the patterns of risk in all of this data and using that to make everything more efficient.
- [05:59]
Mike Linton
That's right. So yeah, again insurers sit in this unique position throughout the supply chain and do in other markets too. But we're focused within supply chain where you're looking across multiple huge swaths of companies. You know, it's one thing to go in and work with a company and say, all right, we're going to look at just your data and we're going to figure out where you might have your risk. That that's all well and good, but we actually, because we work with so many different customers, we can actually find a much wider swath of patterns and then use that to actually obviously we're going to do it, use it to try to build a very profitable organization. But we also want to use that to work with those customers and have them actually benefit from that wider learning that we have.
- [06:44]
Jake Burns
Right. I mean, in a way it's kind of what AWS does with the cloud. Right. It's like a win win. We operate more efficiently, we're able to lower our costs. Right. And then we're able to pass on much of that savings to our customers. So they have lower cost as well. And then we use less energy as a whole as well. So it's kind of, I think oftentimes we fall into these kind of false trade offs where we got to make more profit or we got to help our customers more in a lot of cases you could do and increased sustainability as well. A lot of times all of these things are very linked together. So it's all about increasing efficiency at the end of the day, right?
- [07:21]
Mike Linton
That's exactly right. And it's actually interesting you brought up Even the sustainability and the climate impacts to this. There's actually an interesting stat and I can go back and try to find it, but just in some of these more recent delays that everyone sees in the headlines, whether it's the Baltimore Bridge collapse or the Red Sea blockage, Suez Canal, all these different, you know, major things that are affecting all of us, the amount of CO2 burned because of rerouting because of those is significant and it's something that we don't really think about. I mean, obviously we all want to have our goods arrive and on time and good state and things like that, but the actual climate impacts are very significant. And so that's something that we look at, obviously the delays, the political instability for these different regions. And there's real implications, you know, if, if customers have to reroute through different parts of the world.
- [08:13]
Jake Burns
Yeah. And by using real time data, and I assume using generative AI and some capacity to do this, it really, as those situations change, you could still remain efficient or even get more efficient over time. Maybe you could talk a little bit about how you're using generative AI to accomplish these things.
- [08:29]
Mike Linton
Yeah, absolutely. So we use it in our kind of overall technology data investments, I should say, and generative AI is a big component of it, are broken down into two key areas. One is on the kind of productivity and efficiency gains that we gain from it. And then on the other is predictive risk modeling. That's the two general categories. And we make sure that our investments are built in a very cohesive way. So on the first part, the kind of productivity and efficiency gains, anytime that a customer wants to work with us, there's a submission process and, and there's a ton of unstructured data that comes in and it's a pretty like perfect gen AI use case. All this unstructured data in different formats every time, different people filling it out every time, we send that through LLMs to structure the data. So that way downstream when we are doing that predictive risk modeling, we actually have a foundation of structured data to work with that it also allows us to deliver significantly better service back to our customers. So no one wants to sit around waiting for a quote on their insurance for weeks on end. When we can quickly capture that information, we could feed it into our models and give more of an indicative price. Immediately helps the brokers, it helps the end customers. It also helps us scale our company much more efficiently. So our underwriting staff, our claims professionals, they have a very specific skill and we want them to always focus on what they do best, engaging with the customers, applying their own view of risk, working with the brokers, building relationships. And when we can use these technologies to make them more efficient, it helps them look better, it reduces the admin work that they have to do. Now, like I said, it structures all that data. Then we have separate types of investments on the predictive risk modeling. That includes both LLMs, other types of data science and predictive modeling, things like that. And again, as those get built, then they feed that kind of intake side as well.
- [10:37]
Jake Burns
Can you talk a little bit more about the approach that you've taken with generative AI? What technology you've used, what's worked for you? I know there's a lot of leaders listening to this right now and they're wanting to use generative AI and they're kind of thinking, how do I get started? How did you identify the opportunity? How did you choose which technology to use and what were some of the lessons that you learned?
- [10:56]
Mike Linton
It's very exciting as any type of technologist, it's very exciting to follow and watch. And there's maybe isn't a tendency to try to use all the flashy new things. You know, we do look at that and we want to use that. But at the same time, the best technologists find the right tool for the job, right? And you have to balance obviously the costs, the speed to getting the value, the capabilities within your team, your existing infrastructure, all these different factors go into it. And so we knew that we had a bottleneck in our organization to scale, which was being able to bring on enough customers and assess their risk in the most efficient way. Traditionally, insurance companies grow over a long period of time. And as a startup young company, that's not really something that we have the luxury of. We need to grow quickly. We use Amazon's bedrock because what it really allows us to do, it allows us to prototype quickly as well as move into production very quickly. And what we love about it is it has a number of different mod in it, they continually get updated and frankly, you know, using the most bleeding edge model from a gen perspective, it's not, from a practicality perspective, it's probably not the most efficient way to do it. What's more important is that we can quickly validate a use case and we can move it into production and get the value. That's what I care about. And that might mean using a really advanced model for very specific things. Or more likely it could be using a more tailored model or maybe some of the smaller models out there that that are more efficient and that we can get up and running faster. And so Bedrock specifically was that platform for us. That's what I cared about. When our team evaluated all the different models out there, it was the way that it interacted with our existing infrastructure and the way that we can interact with it immediately, all the tools and infrastructure around it allowed us to actually seize value. So from the start of a conversation that I had with our lead data scientist in last fall, to us gaining real benefits in production was probably three to four months ballpark. And today it's in production, it's scaling, we use it for everything. We're already putting it into other areas of our work streams. So, yeah, it's been a great experience so far.
- [13:13]
Jake Burns
I think you hit the nail on the head in terms of benefits there.
- [13:16]
Mike Linton
Right.
- [13:16]
Jake Burns
Being able to get started very quickly, but still not be kind of locked in on any particular model, being able to choose which model you want to use. And you know, oftentimes, you know, when we choose technology, we don't get it right the first time. I mean, are there any missteps that you made throughout this process? Is there any lessons learned that maybe other folks, you'd say, don't do it this way, learn from my mistakes.
- [13:39]
Mike Linton
Yeah. So kind of similar to what I was saying before, we looked at plenty of providers out there and although it's they have amazing capabilities, the additional overhead and effort they that was going to be involved, like required to use them, which we spent a good month or two trying to use other ones. And we ran into issues again with the infrastructure, with data privacy rights. There's a lot of things that we ran into. And so we kind of took a step back and said, what do we actually need here? What are we actually trying to accomplish? Another example is when we were trying to automate a very specific task. We were applying all kinds of different data science models and we spent good two or three months on it and we couldn't quite get the prediction right. And at the end of the day, we were in about a day's time, we sat down and just wrote, did a basic statistical model and that's all that was needed. So sometimes you want to use the most advanced stuff, sometimes some simple math will do the job.
- [14:38]
Jake Burns
Yeah, no, I could totally empathize with that. Having a technology background, sometimes we have a reputation for wanting to play with the new shiny objects and there's some truth behind that. But the end of the day, when you're running a business, it's about getting value right? And how quickly can I get value? And how efficiently can I get value?
- [14:56]
Podcast Host
We hope you're enjoying this discussion to engage with other business leaders on these topics. Follow us on LinkedIn @AWS Executive Insights. Now back to our conversation.
- [15:09]
Jake Burns
Can you talk a little bit about. Because this is something that I get asked from enterprise leaders all the time about generative AI. How can we ensure that we're going to, this isn't going to be just another shiny object that we're actually going to get business value out of it. What have you learned throughout this process that maybe others can get some value out of in terms of getting value out of using this technology?
- [15:31]
Mike Linton
It's a good question and I think it's something that again, as technologists we have to do with anything that we use, whether it's gen AI or anything. And we have to look at the speed to value with this. And, and so if we can go from an idea to prototype to some type of release in a timeframe that captures the use case, as in, you know, it doesn't pass us by, we can actually, like the customers, can see the benefit. That's what we're looking for. And I think the best way to validate is talk to your users, engage with them, ask them. Did this actually help? Obviously we do all kinds of background analytics and user stats and, and process mapping and, you know, all the techniques that we can employ. But at the end of the day, if the customers aren't actually feeling the benefit of it, it's not really that worth it.
- [16:24]
Jake Burns
So it sounds like this is kind of what I've been hearing a lot from customers who are successful with gen AI and with cloud in general is to kind of run a lot of experiments, but cut off those experiments that aren't providing value very early. Right. Don't chase those things because I think with legacy before cloud, we used to have this huge upfront investment and we'd really, really be motivated to make that work. And so we'd end up kind of really funding projects much longer than we should. Was that something that you noticed as well?
- [16:56]
Mike Linton
Yeah, absolutely. You have to keep asking yourself, why am I doing this? Why am I using this particular technology? Why am I spending something, whatever it might be. Ask yourself that consistently. And if you don't have a good answer, if it's not a clear cut answer, you probably should start asking yourself, maybe I'm going into the deep end here.
- [17:16]
Jake Burns
Yeah. All right. I want to ask you about something very specific that I heard you say fresh frozen food and pharma are hard to insure. Why specifically are they hard to insure? And perhaps more interestingly, how did generative AI help you efficiently insure those types of items?
- [17:33]
Mike Linton
How we kind of characterize our specialty of essential supply chains is if something is fresh or frozen. Otherwise known food and farm are kind of the other way to put it. Right. Sometimes just referred to as cold chain is a different way to put it. That is where our specialty is. The reason those markets are much more difficult to insure is because they have a lot more factors that will ultimately cause the goods to go bad and ultimately trigger a claim. And you know, if you think about moving a chair from one end of the world to the other, yeah, it can get damaged and stolen and those things can all happen. But if you're also moving blueberries from one end of the world to the other, it can get damaged and stolen, but the temperature variation can go. They are much more susceptible to that drop of the box, so to speak. So when those types of risks get introduced, you also have a much shorter time frame. So you know, you need to get those blueberries from one end of the world to the other. Matter of days, maybe weeks, depending on the goods. Whereas again, if a chair sits at a port for an extra three months, there's not going to be too much issue. So when you have that time constraint, you have those environmental conditions that are affecting it all of a sudden. Now you need to figure out what is actually going to cause those temperature variations, what is going to cause those delays and things like that. And so we use Genai and all these other data sources that we bring in to figure out where those patterns are. So if we know that a particular trade route or for example, we know that if there is a trans shipment, so it's going from one boat to the other along the way, you know, that might happen at the Panama Canal or other places around the world that is going to have an increased risk. Because what commonly happens in that case is the container might sit at the port waiting to get loaded onto the next boat for an extra day or two, those goods can go bad. So we know that's like a simple risk factor looking for that transshipment. Well, when we get data coming into us for us to underwrite, sometimes we know that there's going to be a transshipment. Sometimes we have to infer that based on the routes that they're using. Well, to infer that, we need to have good information on those routes. And again, that data might be spread across multiple different documents written in 10 different ways, all these different factors. And so again, using the genai helps us structure all that data to identify those different risk factors.
- [19:58]
Jake Burns
You know, that's really interesting. You're a startup, right? You're not, you're competing with very big insurance companies here, if I understand correctly. And you're able to do the work of a very large insurer with a very lean, small team. Now that's incredible, right, for an inspirational story for any startup. But it's also, I would say, you know, I run into sometimes when I'm engaging with these large companies, not necessarily insurance companies, but large established companies, and they say, you know, we're not really that excited about Generative AI because we're doing well right now. Right. We're on a great trajectory. But what I always say is there's this hypothetical startup that you may not be considering that's going to come and disrupt your industry. Well, I just happen to be talking to the co founder of one such startup. So let this be a warning of sorts. You can't really rest on your laurels. I mean, this technology can take you from nothing to a very viable, successful company like it did for you, but it could also disrupt an industry and cause these kind of established players to have to rethink the way they do things. Are you aware of any of these large organizations kind of, you know, taking your lead and adopting this technology?
- [21:10]
Mike Linton
Yeah, it's, it's actually interesting because that exact scenario is what I say to my team. I say, even as us, as young startup, you were going to go out and beat us right? Now, what would you do? Right? And just, and to kind of keep that, that idea fresh in our mind. I'm asking myself that every year. And if I was going to go out and compete with myself, I would be making these investments, right? You know, as a, as a young company, we can't go hire 100 more people on a whim, right? Like, but we sometimes need to do the work of 100 more people. Well, so where do we get the efficiency gains? Well, how do we make our team operate two times, three times, four times more efficient? And when you look at, at least within insurance operations, the, the amount of overhead and admin work is very significant. And that's obviously the most ripe area for us is like, okay, well we're still going to bring on the best underwriters, we're still going to bring on the best claims people, like they have a real skill to bring for us, but I want to make them as efficient as possible. I want to get the most out of them we can. Now as far as, you know, other insurance companies out there, yeah, you're right. They're looking at it. They obviously make technology investments. I'm sure they're making some similar ones. But I think the things that are maybe more inhibiting one, when you are an established company and this is kind of the standard innovators dilemma, there's an inherent weight to your operations and your technology and your infrastructure that prevent you from doing this. So yes, we're able to quick test out a use case and use a new piece of technology and pull in a new model and try it out and then ditch it the next week. To just go through that as a larger company, a more established company, they're gonna, they're just inherently gonna be slower. The other area is we are pretty, I would say innovative with the way that we're going to get other data that ultimately feed these models. So customers are collecting a lot of data on their supply chains already. It's not like they're just sitting there just waiting for someone to tell them what to do. They're collecting a lot. They, they know more about their supply chains than anyone and they have struggled to communicate with their insurers in a way that is effective. So they might know that actually I'm not that bad of a risk. I know that I ship X commodity and you think that this is a bad commodity, but look at all the things I do and sometimes that can't be communicated in a kind of traditional submission file. And so we are very open to working with customers in a different way if they have a different type of data that they want to share with us and say, hey, here's what it looks like. A couple months ago we launched, launched our data partner program with all the leading sensor companies in the world. So that way a customer can say, hey, I work with this sensor company to monitor the temperature in all my goods. You can go pull that data from them automatically and take a look at our risk. And so there's other areas besides just specific technology investments to work with customers to kind of get that better view of risk.
- [24:05]
Jake Burns
Yeah, yeah. I think there's a lot of lessons that large companies can learn from startups such as yourself. And I think that, you know, you make some very valid points as to some of the challenges that have, but I think a lot of those are kind of self inflicted challenges, you know what I mean? I think really learning how to be innovative within those kind of constraints. I mean they have a lot of advantages as well. They have more resources. Right. You mentioned hiring 100 employees. These large companies can very easily hire 100 employees. If they were to take the same mindset and use the same technology you're using, in theory, they would be able to be able to kind of be as agile and maybe even at a larger scale. So I think there's a lot of lessons to be learned here. There's one last quote that I heard from you that I want to mention because I thought it was just, it was beautiful. You're giving your customers a discount if they use better practices. And that is one of the most effective ways to create change. I thought that was fascinating because we kind of touched upon this a little bit earlier. But you're not only rewarding your customers for being more efficient, but you're actually shaping their behavior to be more efficient as well. And I thought that that was really, really interesting. You think this a concept that could be extended to other industries besides insurance?
- [25:14]
Mike Linton
Yeah, I, I, I certainly believe so, and it has in other areas, but I think it can be used in a much wider sense. It's a very central part to our kind of ethos and how we want to operate. You know, our goal is not to just build an insurance company. Again, that's not what we set out to do. Our goal was to drive some type of impact in this area. And one of the most effective ways to drive change is through the use of incentives. And I actually learned this as I have come into the insurance industry and I really have grown an appreciation for the power, the incentives that are at play now. It's not to say that insurance companies are always using these incentives, but they are in a privileged position to do so. And I think most of them want to. You know, the people that we work with in the insurance industry, they want the to to do this in one shape form or another. Actually, historically there is some kind of interesting examples. So, you know, when, for example, fire risk and workers comp risk and all these other things, when insurers started to actually take, take the risk, then they actually were like, oh, actually how do we actually reduce this? So that's when they started mandating that you got to have a fire extinguisher or you got to use certain building materials or whatever it might be, there's historical risk and you can watch it over decades. The amount of building fires that were happening and the amount of type of claims that were happening, like it really does effectively work. And so, but I don't think it's been done very well within the supply chain. And that's what we want to do. And so customers and humans in general respond to incentives. That's, that's, you know, generally how we operate. And so that's what we want to employ here. You know, customers obviously want to run an effective, profitable organization. And so when we can show that they're going to have, you know, less risk, less waste, less things go bad by doing certain practices and then obviously they're going to save on their cargo premiums, that's a pretty good incentive.
- [27:07]
Jake Burns
Yeah. What I, what I think is key here is it's not just like a one way thing. It's like if you underwrite this and give a better price because they're doing certain things, you're actually providing the transparency as to why. And I think that's the key. Right? That's that, that's what's giving your customers the ability to change their behavior for the better. Right?
- [27:25]
Mike Linton
That's exactly right. Yes.
- [27:27]
Jake Burns
So, okay, we have a lot of leaders who listen to this from various industries. Say you have an enterprise CIO sitting in front of you who says, I want to take advantage of generative AI. What would be the kind of the biggest lesson or the biggest piece of advice that you would give them?
- [27:40]
Mike Linton
I think the biggest piece of advice is don't get caught up in the hype and think very practically around things. Don't feel like you need to use the latest. The model that came out a year ago is probably going to work just fine for you. You. There's only a handful of organizations that probably need the most bleeding edge thing. And I think we can get all caught up in saying, well, I gotta use the latest version of this model or from this provider. For most business tasks out there, you will gain significant margins by using probably just about any of them. And so when you think about it like that, what then is the most important is how fast you get that value. What is the fastest way for Bee to gain that value? And so just be very practical about what you're looking at. You can go and have fun on the side using the latest models, doing all kinds of fun tricks and things. It's great and I love doing them as well. But at the end of the day, we're here to be technology professionals and be efficient with the use of our technology.
- [28:40]
Jake Burns
Yeah, I think that's great advice. It's really kind of like Pareto principle, right? You could use a model that is 5% less accurate and get maybe 10x time to value. And that time to value is going to be so much more important than that. 5%, right?
- [28:56]
Mike Linton
That's exactly right. Yep.
- [28:57]
Jake Burns
All right, so lastly, let me ask you, what's been the most rewarding and impactful part of your work at Parcel so far?
- [29:03]
Mike Linton
That's an interesting question. Parcel has been an incredibly rewarding experience for me in my career, but it's evolved. So as I briefly mentioned, we, we started within global health and, and vaccine delivery and improving the, the quality of those goods. That's a very like in your face impact. Like when you know that you protected certain vaccines in a remote health facility in Africa where they've never had the view of if their vaccines are good or not, they just have to give them out either way. Like, that's a very in your face impact. And that's been incredibly rewarding. Where we are today, after many evolutions is looked at in a very different way. Yes, it's not as in your face as that, but the converse of that is it's on a much larger scale. And now we are in that position to drive an impact across the globe, across the supply chain, across these commodities, across many more organizations, including vaccines and pharma and all the things that we were doing before and previously, we didn't have that incentive. We didn't have that ability to create that incentive. We could still try to show some of that transparency in terms of where are your risks, but we couldn't do too much about it. Now we have that tool, we have that incentive to actually affect that change on a much larger scale.
- [30:23]
Jake Burns
That's incredible. And what an incredible journey you've been on, an incredible story and what a great success you've had. It's really very inspiring. Mike, I want to thank you so much for joining us today on the podcast. This has been, like I said, really inspiring. So thank you very much.
- [30:38]
Mike Linton
Thanks, Jake. Appreciate you having me.
- [30:41]
Podcast Host
Thanks for listening to this episode of Conversation with Leaders brought to you by AWS Executive Insights. For more on these topics, visit aws.Amazon.com executiveinsights While there, visit our generative AI&ML page for more resources to help you unlock the value of Generative AI for your business.