Transcription:
Bob Burns: (
Hello, everyone. I guess we'll get started here. I think we're a couple minutes sort of past the advertised time here. So thanks everybody for showing up. This session is talking about next level automation and how alternative data, we can talk a little bit about what that means really can be the fuel for enabling automation in the small commercial space. There's a lot of data out there. It's not all kind of created equal, but it has the potential to do a lot of different things, be a source for, answers to questions that you aren't necessarily able to get to today, but also, potentially predictive of a lot of different things, because it's new dimensions on businesses that, you don't necessarily have access to through the, traditional data sources. So we thought that the best way to have this kind of conversation is with one of the, leading, insurers in the small commercial space, the Hartford and we have, Pete hill here today from the Hartford. So I'll let Pete introduce himself. My name by the way is Bob Burns. I'm the Chief Product Officer with Carpe Data.
Pete Hill: (
Hi everyone. I'm Pete Hill. I work at the Hartford, I'm our Vice President of underwriting and I lead our underwriting practice, our appetite, our guidelines, our automated underwriting strategy, our book quality book management, and then our training program as well.
Bob Burns: (
So, I think what we wanna do today is, Pete and I have a little bit of a conversation about this and sort of see where we get and what questions may pop up as a result of that. So maybe Pete, let's just start with how do you characterize where Hartford sits in the small commercial space? What's working, what are your goals? I probably should say and what's your specific charge relative to all that?
Pete Hill: (
Yeah, that's a great start as a level set. So first, like you just mentioned, we view ourselves as one of the market leaders in small commercial business. and in order to keep that position, we need to continue to build market leading products and capabilities. One of the things that we're focusing on is achieving profitable growth and increase and continuing to increase our market, share increasing our customer base. One of the ways we want to do that is continue to become a broader and a deeper risk player. So broader appetite, more classes, more industries, deeper within class and size as well, and continue to grow from there. I would say our goal in general is we want to transform the traditional underwriting process, the question and answer more static process, using data and analytics. So there we know, and where we're working of is the data collection process today is, can be inefficient the questions with an agent directly to a customer, having to go back and forth and answer specific carrier questions that they all differ, how we can transform that process in a fast, easy way is important to us.
Bob Burns: (
So when you talk about transforming is I'm assuming that, there are multiple dimensions to that. One would be gaining greater automation. So efficiency of what of throughput the second piece is, as you say, kind of broadening your appetite, adverse selection must come into play there as well. Can you talk about those things a little bit?
Pete Hill: (
Yeah, absolutely. So from adverse selection standpoint with that versus we just wanna make sure that what we're expecting to get from a customer base, a customer standpoint with our appetite is actually what we're receiving. Being able to identify outliers, being able to identify operations that we may an agent or a customer may think differently than what our carrier perception of a business operation is and being able to make sure that as we use data, we're able to automate make it faster and easier to do business with lower the underwriting touch on our side as well, but make sure that we are still writing the right type of business to maintain that profitable growth in customer base.
Bob Burns: (
So I assume that process starts as, I think it would with most insurers with a broad level classification of the business out of the box right. Get an application in. First thing you do is classify, and then you might kind of hone that down a little bit to start talking about underwriting matching, but on our underwrite appetite matching, but in terms of the, classification process, what sort of current state, and what do you see the role of, alternative data, playing in that realm?
Pete Hill: (
Yeah. So first, like you just said, classification's critical. That is the first step in the process from appetite rating coverage needs determining an accurate premium. That is the starting point in everything. We need to make sure that, like you just set around adverse selection. If we have the wrong classification, if we get incorrect data, you know, we could be off base for everything that we're trying to do to protect a customer, provide them peace of mind with their insurance while they go run their small business. So we certainly need to start off on the right path with accurate data to be able to classify broadly their operations and then more specifically within classes, you know, what their needs may be. We have put a lot of focus on classification. We have spent a lot of time trying to get that starting point accurate we've, built capabilities to help agents and customers, both suggest classes help with, you know, their own classification.
Pete Hill: (
As I mentioned, they work across multiple different carriers who have different classes, different definitions, different terms need to make sure that it they're aligned and can get whatever assistance we can provide from a classification standpoint, the kind of the classification help or the classification gap that we work through in that as well as being a broad national carrier, we from a data standpoint need to make sure we can get the right data across different states, across different industries, like I mentioned, across different sizes. So that becomes a big piece of where we can help bring in that process, across different state lines and different classification terminologies. And then from a class standpoint, where we'd struggle as well to your question around the gap is a lot of the customers in small business don't have employees. A lot of them do some of their jobs is hobbies. Some of it's is home based business. So we need to make sure that we can also use that alternative data use, you know, web data at times aggregated data, and be able to find those customers who may not show up in traditional databases, but also have some social media footprint out there.
Bob Burns: (
So it's so part of it's finding it, and then the second part is how do I have enough data, even if I found it to say, I'm not coming to you necessarily with a class and you're not buying it from sort of traditional class provider. Correct.
Pete Hill: (
Exactly, so there's what 33 more than 33 million small businesses in the US. So being able to have across that range of larger end of our small business, all the way down to your micro customers and is important to get data, not only when an agent or customer is looking for help, but also when we're trying to manage our own book as well.
Bob Burns: (
Let's assume you've got something, with a classification that, you're comfortable with, and I'm assuming by the way that we're talking, you may try to sort of standardize some class, but ultimately you've got some fairly detailed proprietary mapping of how you think what this business really is and where it should set. Yes so, and that alternative data is really being fed into your model and providing sort of some increased level of accuracy there.
Pete Hill: (
Yeah. So that's how we're working from a class standpoint is bringing the data in categorizing at the highest level, being able to drill down from there and being able to take in off of minimal information, how we can predictively quickly, easily, accurately classify customers and start to move through the underwriting process.
Bob Burns: (
So let's assume you've sort of classified it correctly as you sort of define that. The next piece I assume is maybe you can talk about this a little bit is matching appetite on more nuanced underwriting preferences. So, one, I think we've talked about a little bit in the past is, I'm a landscaper, there's no question. I'm a landscaper, but I may be cutting larger trees down versus doing some lighter trimming, that sort of thing. Yeah, you talk about what you've seen in terms of alternative data in those sorts of, examples.
Pete Hill: (
Sure. We have really enjoyed using that example with you guys over time. So, but that's the one that, sticks, but it's absolutely true. So we just talked about from a classification standpoint, you start at the highest level where there are certain classes, you mentioned the landscaper, but they're not all the same. So like I had said before, when you start to think about rating, when you start to think about coverage needs for that business owner to be adequately protected, there's a lot of nuance within there. There's a lot of differences that we'd also have from an underwriting standpoint, we are not fully automated. So there's still there's business that needs to go to an underwriter and rightfully so should be, looked at for deeper review in those pockets when the data tells us that it could be. So, to your point, there's some landscapers that may only mow lawns. There are some that may, you know, do more work at heights or may cut down trees. We wanna make sure that we have the right information at our disposal to be able to make those appropriate decisions with the underwriter or the customer and the agent.
Bob Burns: (
So it's a good point that it's so the agents involved there too. So part of this is, presumably giving the agents some or it's improving the agent's experience, you're giving them some information that they can act on. Is that a good way to sort of think about using that data as opposed to saying, well, here's a landscape and this is exactly what they do versus? I'm gonna sort of lead, whoever the applicant is through a process that says, it looks like you do this, therefore, give you some more information on it.
Pete Hill: (
So we do some of both, but it is definitely helping lead through that process. We wanna be easy to do business with. We wanna make sure that if we have information, we can make the right decision for us at the end of the day, but we can also make the right coverage recommendations. We can help with, underwriting questions in that process overall on things that we know about a customer and get those to the right places from an underwriting review standpoint, as we know that they should not all just go through one static process for every class, every customer, every state, every line of business, because the businesses, as we get smarter and used alternative data, we've got different signals on customers that can help inform level underwriting intensity required.
Bob Burns: (
So that means that you're really sort of consuming this data real time in the process of data coming kind of into the application being submitted or through whatever the mechanism is that.
Pete Hill: (
Exactly. So on our, particularly in our customer acquisition space, our new customer space, you know, we've obviously got processes in place on the renewal side of things as our book management, but in the new customer space where we see a customer for the first time that need and where we've partnered together too, is on that, those quick responses of data and a quick, easy, consumable manner that can help be input into our underwriting process to help us make smarter, faster decisions.
Bob Burns: (
So, yeah, so that's kind of where I'm going with the question in some ways is part of the challenge I think is that, we've heard from you is that, it's gotta be fast, right? If I want to use it in the manner that you're talking about, it has to perform in sort of a sub second manner, but it has to be in some form that is not too raw, right. It's gotta be the right product wrapping, if you will, of whatever, the data point that you're talking about.
Pete Hill: (
Right! Yeah. So exactly right. We don't want an agent, a customer coming to us and, you know, starting to give us some information, we try to use data and analytics to help evolve that traditional process I referenced earlier, and then they're waiting for 30 seconds or a minute for a response back to get to the quote to get to the screen. So that instantaneous response in a consumable way, like you just mentioned, that gives us the right level of signal into being able to make a decision on that. Like you mentioned a class, a question, a premium, those various components that we can not have to put in front of an underwriter and a human and say, take a look at these and then make an assessment. How do we continue to do that in an automated way? And that's where the data has come in and been most helpful in the way that you've structured it.
Bob Burns: (
The automation in this case could be sort of described as sort of two things, one answering the question, so you don't have to ask too much, right. You're leading somebody through it. So you're improving an experience and there's not as many clicks through screens, etc. But then there's another element I'm assuming of, I'm routing this where I need to route it, right. If somebody needs to touch this, I get it as fast as possible to that person. Is that a fair characterization?
Pete Hill: (
Yeah, absolutely. So to explain that process a little bit more, I guess, around the question or the touch is there's two things we can do there's rating questions, right. That help determine your premium, like sales volumes and things like that and then there's underwriting questions that help get at the appetite we mentioned earlier, are you within our risk profile? Are you not, you know, what coverage needs would you have that get at? Should we, and would we want to collectively provide insurance for that customer at all or not regardless of premium. So there's two different buckets there and the faster and the easier that we can do that. And come to that end state is great, right from this, the straight through processing, the quote, throughput of ease, but then on the back end from a question standpoint, we've got options. If you know something from data, do you want to prefill that answer and the information for the agent of the customer, if you know the answer to it, particularly non rating questions, would you want to, do you need to ask it at all? So being able to look across the entire process, there is a little bit of what I meant by being able to transform that process more dynamically.
Bob Burns: (
Yeah. Makes sense. So when you look at having additional data points like that, and potentially the using alternative data answer questions that, you know, like it's hard enough to answer the questions that are on your application today, are there new things that are potentially out there? How much have you evaluated that? And do you see a lot of potential for saying, look, you know, I've always asked about X, Y, and Z, you know, maybe whether you have a swimming pool or not, but it turns out there are other things that can potentially be more predictive. and that line up with outcomes a lot better, have you, how much have you looked at that and what's your general sort of sense of the potential there?
Pete Hill: (
Yeah. So we're constantly looking into questions, making sure they're relevant, making sure they're value added, making sure that they would help with a risk decision. I think the other side of it would be that if we are looking at questions and we feel like to your point, are there pools or are there not pools? What do we know that you could make an inference on? You could use data as a proxy to answer another question, even if you don't explicitly state it. And I think that's where some of the data aggregation, like we had mentioned with the indexes and some of the class at a high level, and then getting down to that within a class, what's your level of, complexity, cuz they're not the same. How do you route that quicker and easier to an underwriter to use the data, use the information at their disposal to then make the appropriate underwriting assessment?
Bob Burns: (
Yeah. That makes sense. Are there any sort of, data points that surprised you? We throw some interesting things out there more to stimulate conversations in some cases, whether it's providing hydration stations at a bar or something like that. Have you looked, at some of those sort of emerging data points and how do you think about staying on top of those sorts of things as they develop?
Pete Hill: (
Yeah, so I think the big thing that not a specific question, that we've looked at that adds value, but I think we all know that small business owners never stay the same. So when we look at a customer, particularly at the new business timeframe, we get that as a snapshot and then that customer has their life cycle with us, and are able to change their business. So being able to understand how they've changed over time, maybe even without questions is something that's important that the data can provide to us. When you look at multiple data sources together, multiple attributes together, it helps give you a better best guess of what that customer has been doing, what their operations are, how they're trending and be able to continue to say almost in real time, has this customer shifted and should we look at them differently?
Bob Burns: (
We've been talking mostly about how some of this alternative data answers questions about the business that, help you do things. We talked a little bit about predicting in the sense that you're adding this kind of data to your, your classification prediction, but, can you talk a little bit about other, sort of use cases where you're adding that to models in the underwriting process, just from conceptually.
Pete Hill: (
Yeah, so any place that we can find value in data we've explored and we've tested. I will say we've been on this journey for a while now, but there's still significant opportunity that we can do more, and continue to learn and get better. A big part of that has been the products that we initially use as they get better over time unlock new capabilities. So when you look at, like, I just mentioned that one snapshot in time wanting to be fast, easy, and accurate to get a customer on your books and then make sure that you have the right coverage needs on their backend processes through additional, loss control services on the backend, audit process that we primarily, you know, look at from a comp standpoint and things like that. Then through the underwriting year after year, as they come up for renewal, making sure that we can stay on top of our large enforce book that we have using data and sophistication.
Bob Burns: (
Great. So maybe stepping back here for a moment in terms of, so I keep using the term alternative data, what Carpe data generally means by that is data that's typically not used in traditional or not from traditional sources. So we think about it in, you know, kind of two buckets. One, how can I want to know the question about the landscape or so I might not have had a good source to answer that question in the past, but there's also the notion of these other dimensions of data that are so that can encompass data that we find in more social spots, so feedback on businesses, but also things like, you know, can, can I, understand, all sorts of other, structured or, unstructured data sources that relate to the business that's kind of my definition of alternative data and Carpe definition to it. So it can be lots of different types of records, not just sort of social stuff that we're finding. How do you sort of think about that? And then as you kind of looked in dealing with us, how have you thought about, how do you define that number one and then how do you find somebody that can deal with that kind of data?
Pete Hill: (
Yeah. So the first, I think I agree closely with your assessment of alternative data. I think we generally categorize some things as, you know, credit data or property, data, public data government, as you would name it, those broad buckets. When I think about alternative data and the way that my mind goes as an underwriter is in your personal life, in your business life, the first place that you're probably gonna go to try to find anything, you're planning a vacation, you're looking at a customer is gonna be what's out there on the web, right. What you can find across a variety of sources, because like I mentioned, the micros, a lot of times the micro businesses, newer business ventures, they may not be well established in those traditional bureaus, the traditional sites and so you have to explore elsewhere. I think that's where the alternative data has helped fill those gaps where you may not appear have your own website have any regular traditional bureau, but you're able to still see presence that they have online to advertise their business, customer reviews, that type of information that allows you to learn a whole lot about a business with minimal effort.
Bob Burns: (
That sounds good. So in terms of looking at providers of that, can you talk a little bit about, what works? What doesn't?
Pete Hill: (
Yeah, absolutely.
Bob Burns: (
So how did we get here? Right.
Pete Hill: (
Probably we've enjoyed it. Probably some head slamming on the tables from you guys, but, no we've had a good partnership. The first thing that sticks out as we meet with vendors in general, across a variety of topics, sources, etcetera, is just understanding and being at the table with each other in the honesty of where the product capability is, you can understand pretty quickly what's a concept and what's just a pitch that needs help to be built out versus what's real, right? What's an existing product? What's an existing capability? What is something that can be in production and be delivered? I think that's one of the places that we've had the most success together is you guys came to the table with a real product, real capability, a vision for how you wanted to use it. We went back and forth, as I mentioned before, the speed in which we'd need it, the sub second response time to get that data, the accuracy in which that data would need to be delivered, being able to understand our needs from our business side versus paired with the data capabilities.
Pete Hill: (
I think we've done that extremely well together and understanding both sides of it. I feel like we've made you a little bit of an underwriter along the way with your landscaping risk, but that's, that's really the biggest thing for us is just honesty and is there a capability or do we need to partner together to build something right? What the inputs we would need to provide the vendor, to get a response back in that data. Then over time, just the understanding that we're buying or working with a product, but our expectation continues to be the product will get better over time. The database will grow more customers, more businesses come online, increase the attributes that are within the data as well, as well as continue to invest in new products that we can use the same partner, the same vendor for on multiple products. Or you could use that data as a replacement for a different product you have in market, and be able to consolidate those as well from an efficiency and expense standpoint internally.
Bob Burns: (
Great. Well that all sounds great, in terms of how we've sort of progressed down the road together. So I appreciate all that. I maybe we'll open this up a little bit. See if anybody has questions here, we have a few minutes left. We'll take any for either Peter myself in terms of this sort of did Scott.
Audience Member 1: (
So, Peter, can you talk a about sometimes like their main role is to validate data, part of what this helping feed insight can really validate that sometimes we most validate data underwriter has changed?
Pete Hill: (
Yeah, that's a fantastic question and something we spend a lot of time on. So what we think, and what I mentioned before is we want to continue to grow, right? We want to continue to expand our appetite, expand our business footprint as well into where we can provide solutions to the small businesses in the US. So as we continue to automate, we are not discrediting or taking away the value of the underwriter at all. We are redeploying resources into places that they can provide their expertise and provide their value. So if we wanted to go into a new industry, if we wanted to, as there are emerging trends that are popping up across the US explore, you know, new businesses, new ventures in general, or how do we get that business instead of data validation. Say, there's something here, I think to Bob's point earlier, there's something here, but we don't know exactly what it is. Making sure that we're getting that to the underwriter to spend their time, make the appropriate decision from an underwriting in a pricing and rating standpoint, instead of static rules by class, by state, where they're touching all of the same business for all the same reasons every time.
Pete Hill: (
And it gets a little into that repetitive, just purely data validation. So there is a skill build. There is a training component that comes with that, but I would say there's general excitement over how the underwriting role has changed with what they're experiencing and what they're looking at.
Bob Burns: (
And you guys have very specific sort of targets right. Of how much you don't touch. Right? So, so your underwriters are presumably kind of embracing that, general notion and, and effectively saying, you know, I don't have to touch the stuff, they don't need to touch. So that's actually improves the job is effectively what you're saying.
Pete Hill: (
So there's, there's partnership there. There's, there's excitement there. Being an underwriter myself, I was always excited to work on some of the largest, most complex things to generate the premium. So if we can work through that space, but we can also automate some of the lower exposure pieces of it and remove that from their desk, then that's certainly been well received as well across both our new business and our renewal side of the book.
Pete Hill: (
Yeah. So not specifics as far as a number goes, I would say we continue to look at what our own internal capabilities are, how far we can expand. We wanna make sure we are not deteriorating any of the underwriting integrity in the decisions just for taking that for automation. So we don't wanna just open the floodgates and let business through and make sure that we don't ever kind of, get complacent in the accuracy of the data either. So we do want the products to get better. What I mean by that mostly is how they can, how we can continue to increase the number of customers as we kind of call it a hit rate, the number of customers, we can send multiple data sources, card pay and get response back on data versus what comes back. and we, don't get data on or minimal data.
Pete Hill: (
We go through more of the traditional place as well. So really that's the evolution of the product that they would build out across more industries, more business profiles in the US. Then a lot of that, we're continuing to try to be a market leader but as the market changes, as everyone, I'm not alone at the Hartford, that's trying to do this from an automated standpoint, by any means. So as we see the market shifting that kind of plays into the areas that would go as well.
Bob Burns: (
I just add sort of from the provider side, that the way we look at it the sort of two issues here. One is as Pete sort of said, is, can we find the business right? The more particularly in small commercial as you go to some of those individual, I'm mowing lawns kind of thing, right? Those are harder and harder to find, right. Once you find them, then question is data sufficiency. So as the products that, the Hartford pushes on to say, "Hey, we need to understand these things." Data sufficiency becomes a big issue as well. So, constantly sort of pursuing sources to be able to answer those things. Our number of sources, we look at grows and grows and grows to be able to sort of be responsive to that as well. But we have to balance those things. Two things of, I could find it, but finding it isn't enough. I gotta have really sufficiency there. That's right.
Pete Hill: (
My jobs to tell them how.
Pete Hill: (
We need. Yeah, exactly. Right.
Bob Burns: (
I need to go find where that is.
Pete Hill: (
Exactly.
Bob Burns: (
I mean, we, cuz there's a whole bunch of, yeah.
Pete Hill: (
I think we had a question in the back too.
Audience Member 1: (
So I think that automation...?
Pete Hill: (
Yeah, that's a great question. So just for the recording, a lot of it is, you know, if you're gonna trust, make sure I get it right. If we're gonna trust the data, are you able to stand by those decisions and what is kind of the ramification or repercussion if you had to go back on that, is that fair for a partnership or a trust standpoint with a customer or an agent.
Pete Hill: (
We're not perfect. I'm not gonna pretend that we're perfect in those decisions. I have a lot of confidence in the outcomes that we make our ability to stand by those and a lot of that is the data validation that we do up front. We've worked with vendors, we've worked with Bob and team for a long period of time, testing out the accuracy of that information. We fortunately, for small commercial, we've been doing this for over 30 years. We've got a fair amount of our own internal data to help do some comparison. So we, aren't just having to rely solely on the vendor telling us this is accurate. We can do a lot of those reviews so that when we wanna put something in market, when we wanna adjust maybe a question, a rule to Bob's point, we feel like we know what we're getting. We feel like we have the right backend checks to maintain that. We're not taking on any unnecessary risk. I would say for the most part, well I started with, we're not perfect. The feedback we've received from our brokers, our agents, our customers of the speed, the ease, the automation has been well received and far, you know, kind of outweighed the one off or a couple of instances where we've had to change the decision.
Bob Burns: (
Yeah. I would just add, from the provider side on that, you know, one of the ways we try to make that work for Pete is to bring a variety of tools to answering any specific questions. So if we're trying to understand, you know, the landscaper question or even hours of operation, we're looking at lots of different sources for the answer to that we're evaluating which sources we agree with in which case. We look at when we combine, you know, looking at structured data, unstructured data, models, etcetera, to try to say how many things do we bring together to answer this question? So it's not, do you believe this source or not? It's we're using our evaluation of all those things to help and again it's often not perfect, but the more we're running through these things, the more accuracy we have, and I think it's being worn out by experience over time here. Great question. So I think we're up against time. So, I appreciate everybody, joining and, any other questions feel free to see us afterwards.